An Integrated Conceptual Framework for Low-Carbon and Cost-Effective Building Design Optimisation

Abstract

Higher construction costs (CCs) linked to carbon reduction methods have hindered the adoption of low-carbon approaches in the built environment. The simultaneous minimisation of upfront embodied carbon (EC) and CCs has not received much attention in building design optimisation (BDO) research; most studies prioritise operational energy, operational carbon, and operational cost reduction. This paper develops an integrated conceptual framework for low-carbon, cost-effective BDO, particularly targeting upfront EC and CCs, to fill this research gap and meet industry demands. A systematic literature review was conducted following PRISMA guidelines, synthesising 41 peer-reviewed articles published between 2015 and 2026. Thematic and content analyses were employed to extract and categorise key methodological components, including optimisation problem characterisation, objective-driven design variable selection, constraint modelling, algorithm selection, and evaluation and validation approaches. Subsequently, the developed conceptual framework was validated through semi-structured expert interviews with participants comprising BDO researchers and building designers in the construction field. A cross-mapping of optimisation objectives, optimised parameters, and design variables was developed to clarify their interrelationships, alongside structured criteria for optimisation algorithm selection. Based on these insights, a conceptual framework named “ICCO-BD (Integrated Upfront Carbon and Construction Cost Optimisation for Building Design) framework” is proposed and validated, integrating problem formulation, parametric modelling, multi-objective optimisation, and systematic Pareto-based evaluation into a coherent end-to-end workflow, enabling improved time efficiency through reduced redesign iterations, enhanced solution quality via better pareto front exploration, and more robust decision-making through clearer trade-off interpretation. While expert feedback indicated strong conceptual relevance and practical applicability, the framework remains conceptual in nature and requires further empirical verification through real-world case studies and optimisation applications before broader industry implementation.

1. Introduction

The rapid evolution of digital technologies has significantly transformed architectural practice and reshaped methodologies within the built environment sector. Computational design, parametric modelling, and simulation-driven workflows have shifted building design from intuition-based approaches toward data-driven computational environments where multiple performance criteria can be simultaneously evaluated and optimised [1].

Despite these technological advancements, the built environment remains a major contributor to global environmental degradation. Buildings account for nearly 40% of energy-related CO₂ emissions [2], underscoring their substantial role in climate change. Given that operational carbon (OC) dominates building-related emissions, research efforts have predominantly focused on reducing operational impacts [3]. Consequently, it is apparent that comparatively limited attention has been directed toward upfront EC emissions associated with the pre-use phases (A1–A5) of buildings [4]. Unlike operational emissions, which can be reduced through retrofits and performance upgrades, upfront EC becomes irreversibly embedded once construction is completed [5], thereby significantly constraining opportunities for subsequent mitigation. Therefore, proactive mitigation of upfront EC during the early design phase is paramount to achieving long-term climate goals within the built environment.

Simultaneously, financial constraints pose significant barriers to the widespread adoption of low-carbon design strategies, as sustainable building solutions are often perceived as cost-intensive [6]. Consequently, environmental impact minimisation and CC reduction emerge as dual, and often conflicting, objectives in building projects [7]. Addressing these objectives concurrently requires systematic and rigorous methodological approaches.

The building design phase offers the greatest leverage for influencing both carbon emissions and CCs, as conceptual design decisions determine material selection, structural configuration, geometry, and system integration parameters that significantly affect environmental and economic performance [8]. Hence, early-stage decision-making is central to achieving low-carbon and cost-effective buildings. Accordingly, BDO provides a structured computational framework [9] for balancing both upfront EC and CCs of building designs. Since carbon and cost are conflicting objectives, multi-objective optimisation (MOO) techniques become essential to generate Pareto-optimal trade-off solutions in BDO [10]. Optimisation in building design is not a standalone task; rather, it comprises an interconnected sequence of interdependent processes. According to Zeng et al. [11], BDO comprises three phases, including problem formulation, algorithmic solution search, and performance evaluation. The same author discovered that the algorithmic solution search phase has been more intensively discussed in the literature than the other two phases, revealing a prevailing research gap in the comprehensive evaluation of the entire BDO process.

Although optimisation techniques have been widely applied in building design research, most existing studies primarily emphasise OC or lifecycle energy performance. Limited research has systematically integrated CCs with upfront EC, particularly within the pre-use stages (A1–A5). Where EC is addressed, system boundaries are often confined to cradle-to-gate assessments due to data limitations, thereby overlooking transportation and on-site construction impacts. Furthermore, the selection of optimisation algorithms, an element that fundamentally determines robustness, convergence behaviour, and quality of generated solutions, remains underexplored. Given that algorithm suitability is inherently problem-dependent, varying according to objective conflicts, design variable characteristics, and constraint structures [11], the absence of structured selection criteria represents a significant methodological gap. This limitation is particularly critical as computational optimisation applications are still emerging within built environment research and practice.

Accordingly, this study aims to develop a conceptual, integrated four-phase framework for low-carbon, cost-effective BDO, with explicit emphasis on upfront EC and CCs throughout the entire optimisation process. This study is guided by two primary objectives: (i) to identify, map, and analyse existing domain knowledge on carbon- and cost-focused BDO, including optimisation problem formulation, design variables and objective functions and respective constraints considered, optimisation strategies and algorithms employed, post-optimisation result evaluation and validation methods, and challenges encountered in BDO process and, (ii) based on insights derived from the systematic literature review, to develop a detailed theoretical framework that comprehensively discusses each phase of the upfront EC and CC-oriented BDO process.

The novelty of this research lies in its holistic integration of upfront EC and CCs within early-stage optimisation, coupled with a structured algorithm selection guideline and an end-to-end process extending from problem formulation to an actionable design decision support framework.

2. Materials and Methods

To develop the conceptual framework mentioned above, this study systematically reviewed 41 peer-reviewed articles to synthesise current research on EC and cost-focused BDO in stage 1. The conceptual framework was developed based on the findings from the systematic literature review and subsequently evaluated through six interviews in stage 2.

2.1. Stage 1: Systematic Literature Review

The systematic literature review (SLR) was conducted using PRISMA to identify, select, and review existing research in a structured and thorough manner. Employing a systematic approach ensures that the review is transparent and reliable [12]. The methodology of the study for the SLR was formulated by defining the research questions, selecting relevant studies, summarising the collected information, and synthesising and interpreting the findings [13]. This study develops a conceptual framework for modelling upfront EC and cost-driven BDO. The research questions were formulated to systematically examine how optimisation problems and corresponding models are developed in existing studies. Accordingly, six research questions are structured around the key phases of the BDO process.

• RQ1: How are optimisation problems formulated, and how are carbon and cost objective functions integrated within BDO studies?
• RQ2: What design variables and constraints are adopted in optimisation-based building design studies, and how do they influence carbon–cost trade-offs?
• RQ3: What end-to-end optimisation workflows, tools, and simulation environments are used to support carbon- and cost-driven building design optimisation?
• RQ4: What optimisation strategies are employed in BDO, and what criteria are adopted for selecting suitable optimisation algorithms?
• RQ5: How are Pareto-optimal solutions evaluated, validated, and translated into decision-support outcomes?
• RQ6: What methodological challenges are encountered in BDO studies?

Articles were retrieved from the Scopus database due to its extensive coverage [14] and recognition as one of the most comprehensive databases for peer-reviewed scholarly publications [15]. Compared with Web of Science, Scopus indexes a larger number of journals across the engineering and built environment disciplines and offers a relatively faster indexing process, enabling the inclusion of recent studies [16]. Although relying on a single database may risk omitting studies indexed elsewhere, it simplifies data management [17]. Moreover, the initial search retrieved over 900 records, providing a sufficiently large dataset for the systematic literature review. Moreover, several recent systematic literature reviews conducted within the built environment domain [18,19,20] have relied exclusively on the Scopus database, demonstrating its suitability and comprehensiveness as a standalone source for conducting rigorous and systematic literature reviews. Accordingly, the Scopus database search was conducted on 15 January 2026. The search strategy was developed using Boolean logic and applied to the TITLE–ABS–KEY fields of the Scopus database. The following search string was used: (“multi objective” OR “multi-objective”) AND (Optimiz* OR Optimis*) AND (“Building Design” OR “Architectural Design” OR design*) AND (“Embodied Carbon” OR “ECE” OR “LCCE” OR “life cycle carbon emission” OR Construction cost” OR “life cycle cost” OR “LCC” OR cost OR carbon) AND (Building* OR hous*). The search was limited to the Engineering subject area as defined by the Scopus subject classification system. This “Engineering” filter includes publications categorised under engineering-related fields, such as civil engineering, structural engineering, construction engineering, etc., while excluding studies classified solely under other domains, such as Environmental Science, Medical Science, Material Science, Mathematics, or Social Sciences. Although related disciplines such as Environmental Science and Energy contribute to sustainability research, the core focus of this study lies in engineering-driven design decision-making and optimisation of building designs, making Engineering the most relevant and appropriate field. The search was restricted to English-language publications published between 2015 and 2026, ensuring a concentrated focus on recent research endeavours [21]. The search results were also restricted to peer-reviewed journal articles and conference papers to ensure the inclusion of high-quality scholarly sources, while books and review papers were excluded. According to Paul et al. [22], books and book chapters are less favoured in SLR studies as they are less likely to contribute to scholarly developments. Based on that, books generally offer less detailed reporting of specific optimisation workflows, algorithms, and validation procedures required for the objectives of this review. Furthermore, review articles were excluded to avoid potential duplication of evidence, as their findings are derived from previously published primary studies [23]. Excluding review papers also helped prevent the repeated inclusion of similar themes and ensured that the analysis was based solely on original research, providing first-hand methodological and empirical contributions [21]. In addition, only articles with a final publication status were included. Papers labelled as “Article in Press” were excluded because their content or bibliographic details may change before official publication. Restricting the dataset to finally published articles ensures that all selected studies have completed the peer-review process and provides stable and traceable references for the systematic literature review.

To ensure the reliability of the evidence base, only blind peer-reviewed journal articles and conference proceedings indexed in Scopus were included in the review. Furthermore, during the analysis stage, the methodological robustness of each study was examined by considering aspects such as the clarity of research objectives, the appropriateness of optimisation methodologies, the transparency of modelling approaches, and the presence of procedures for results evaluation and validation.

As illustrated in Figure 1, an initial pool of 920 articles was identified through the database search. Following the title and abstract screening, 793 articles were excluded, while five articles could not be retrieved for full-text review. Subsequently, 81 articles were excluded during the full-text screening stage, resulting in 41 articles being retained for in-depth qualitative analysis. To minimise subjectivity during the study selection process, a predefined article exclusion criterion (

Table 1. Exclusion criteria for the screening stages.

Stage	Exclusion Criteria
Searching	Books, review papers, and editorials Articles in press Articles not in the Engineering domain Articles not published in the English language Articles not published between 2015 and 2026
Title and Abstract Screening	Articles focused on natural hazards Articles that do not address building design optimisation Articles focusing on material-level studies Articles focusing on industries other than buildings Articles focusing on building retrofitting
Full-Text Screening	Articles that could not be retrieved Articles in which the objective functions do not include embodied carbon reduction or cost minimisation

) was established prior to screening and strictly followed throughout the review. The screening process was conducted in multiple stages, including searching, title and abstract screening, and full-text screening. At each stage, studies were retained or excluded based solely on the predefined eligibility criteria. The screening and eligibility assessments were independently conducted by the first two authors through multiple rounds of review. When ambiguity arose regarding the relevance or eligibility of a study, the article was re-examined and discussed among all authors until consensus was reached, ensuring a consensus-based decision-making process. This iterative cross-validation procedure enhanced consistency in study selection, reduced individual bias, and improved the transparency and reliability of the final dataset. Consequently, the reduction from the initial 920 records to the final sample of 41 studies was guided by a systematic and criteria-driven process rather than subjectively.

The analysis adopted a qualitative synthesis approach guided by the predefined research questions. To systematically extract methodological information from the selected studies, a structured coding framework was employed. Each article was carefully reviewed and coded to identify key methodological attributes, including optimisation objective functions, constraints, design variables, computational tools and software, optimisation algorithms, Pareto-solution evaluation methods, validation approaches, and reported research challenges. The coding process followed an iterative content analysis procedure conceptually similar to open and axial coding used in qualitative research by [24]. In the first stage, relevant elements were extracted from each study through a detailed reading of the full text. In the second stage, similar concepts were grouped into broader analytical categories to ensure consistent classification across the reviewed studies. The coding and data extraction were conducted by the first two authors through multiple rounds of review to ensure accuracy and consistency. Where ambiguity arose in interpreting methodological components or in grouping, the relevant sections of the articles were re-examined and discussed among the other three authors until consensus was reached. The coded information was then organised using thematic analysis and content mapping, enabling cross-study comparison and the identification of recurring methodological patterns [25]. In the final stage, insights derived from the reviewed studies were synthesised to develop an integrated conceptual framework that captures key phases and methodological elements of carbon- and cost-focused BDO.

Although the primary focus of this review is on upfront embodied carbon and construction cost, many building design optimisation studies employ multi-objective frameworks that simultaneously evaluate additional performance indicators such as operational energy use, daylight performance, or thermal comfort. These variables were examined only to understand the broader optimisation context and algorithmic strategies used in BDO studies. However, the synthesis and framework development in this research primarily emphasise design variables and optimisation strategies relevant to EC and CC reduction.

2.2. Stage 2: Conceptual Framework Validity

The ICCO-BD conceptual framework was validated to assess its practical applicability, relevance, and potential for improvement. The validation process involved expert evaluation using a combination of quantitative and qualitative approaches. A total of eight experts specialising in building design optimisation (BDO) and the broader field of building design were selected using a purposive sampling technique, based on their professional experience and domain expertise. During the interview process, all participants evaluated the framework against the established criteria, consistently assigning ratings of either 4 (agree) or 5 (strongly agree), indicating a high level of acceptance. Notably, the data collection process revealed that, after the sixth interview, no substantially new insights emerged that could further enhance the framework. Accordingly, as stated by Saunders et al. [26], in qualitative research, the saturation point is considered as the point at which additional data collection yields minimal or no new information. Furthermore, the same author explained that for qualitative research sample size typically ranges between 5 and 25 participants. Accordingly, it was indicated that theoretical saturation had been reached at the 6th interview. Nevertheless, two additional interviews were conducted to confirm this observation, and the process was concluded after the eighth interview. Although the number of interviewees was limited to eight, the depth and breadth of their expertise ensured a comprehensive evaluation of the framework. All selected participants possessed advanced academic qualifications (Master’s and/or PhD degrees) and substantial experience in BDO, including both professional practice and optimisation-focused research. Their professional experience ranged from 2 to 15 years in building design, with up to 6 years of specialised experience in optimisation research. Moreover, the expert panel represented diverse but complementary perspectives from both academia and industry, including architecture, energy optimisation, cost management, and BDO research. Their collective experience covered a wide range of building types, including residential, commercial, and high-rise developments. As shown in

Table 2. Profile details of the interviewees.

Participants	Position	Type of Organisation	Years of Experience in the Construction Industry	Years of Experience in Building Design	Years of Experience in Optimisation	Education Degree
A	Building Designer	Architecture & Design	10	10	3	BArch, MArch, PhD
B	Architect	Architecture & Design	6	6	5	BArch, MSc
C	Researcher in building energy optimisation	Tertiary Education	8	8	6	BSc, PhD
D	Research expert in BDO	Tertiary Education	2	2	2	BSc, MSc, PhD
E	Architect	Architecture & Design	15	15	3	BArch, MSc
F	Researcher in building energy optimisation	Tertiary Education	7	7	4	BSc, MSc, PhD
G	Cost Manager in the design team	Contractor	5	5	2	BSc, PhD
H	Researcher in BDO	Tertiary Education	8	5	3	BArch, March, PhD

, four participants are academic researchers in BDO, while the remaining four are practising building designers with strong architectural backgrounds. This balanced representation further strengthens the validity of the evaluation. Importantly, a high degree of convergence was observed across both quantitative ratings and qualitative feedback. The consistency of responses indicates strong agreement among participants, reinforcing the credibility and robustness of the framework. The adequacy of the sample size is also supported by prior studies in a similar field. For example, Tong et al. [20] validated their conceptual framework using five expert interviews, while Vergara et al. [27] conducted validation with four experts. In comparison, the inclusion of eight experts in this study provides a sufficiently rigorous basis for evaluation.

The framework was presented to the experts and evaluated using a structured questionnaire. Participants assessed the framework in terms of its structure, content, process, and feasibility for adoption in achieving low-carbon and cost-effective building designs. A five-point Likert scale was employed (1 = strongly disagree, 2 = disagree, 3 = neutral, 4 = agree, and 5 = strongly agree). In addition, semi-structured interviews were conducted to capture in-depth insights and recommendations for improvement. The collected data were analysed using a mixed-method approach. Quantitative responses were analysed through mean, standard deviation (SD) and interquartile range (IQR) score comparisons, while qualitative interview data were examined using content analysis.

The validity of the proposed framework was ensured through face and content validity. Face validity was established through expert assessment of the clarity, structure, and coherence of the framework. Content validity was primarily derived from the SLR, which informed the identification and integration of key components of BDO, including objective functions, constraints, design variables, optimisation algorithms, and evaluation methods, ensuring a strong theoretical foundation. This was further reinforced through expert evaluation, where participants assessed the completeness and relevance of the framework and identified any missing or redundant elements. Furthermore, integrating Likert-scale evaluation with semi-structured interviews enabled methodological triangulation, thereby enhancing the robustness, credibility, and practical relevance of the framework.

3. Results

3.1. Concept of the Study

Carbon and cost-dominated building design optimisation research demonstrates distinct conceptual orientations, reflecting differences in research intent. To systematically interpret these differences, the reviewed studies were classified according to their primary research aim and dominant contribution. Based on this analysis, five key conceptual categories were identified: optimisation framework development, decision-support tool development, optimisation model development, optimisation methodology development, and performance assessment of optimisation approaches (Table 3).

A substantial portion of the literature focuses on optimisation framework development (47%), emphasising a strong research trend toward structuring and formalising the optimisation process [11]. The studies focused on optimisation framework development propose integrated, multi-stage workflows to adopt in early-stage design, where design flexibility is high, and decisions have a disproportionate impact on environmental and economic performance [2,8,28].

The optimisation methodology development category comprises studies that advance the optimisation approach itself, introducing novel or hybrid methods to improve solution quality, robustness, or efficiency in complex multi-objective design problems [29].

Meanwhile, optimisation model development studies focus on constructing surrogate models based on machine-learning techniques, to approximate building performance with reduced computational demand [30]. These models are commonly embedded within surrogate-assisted optimisation workflows to support early-stage design under limited data and time constraints [8,31].

Studies classified under performance assessment (12%) primarily aim to evaluate the outcomes of design decisions or the effectiveness of optimisation algorithms. Rather than proposing new frameworks, these studies focus on quantifying performance impacts [32] or benchmarking optimisation algorithms in terms of convergence, solution diversity, or computational efficiency [33].

Decision-support tool development studies emphasise translating optimisation results into actionable and interpretable outputs for designers and stakeholders [34]. These studies prioritise presenting trade-offs, Pareto-optimal solutions, and life-cycle performance insights to facilitate informed decision-making during early design stages [35].

Table 3. Classification of reviewed studies according to the research concept.

Category	Description	References
Optimisation Framework Development	Studies proposing structured optimisation workflows integrating modelling, optimisation, and evaluation	[1,2,3,6,8,9,10,11,28,31,36,37,38,39,40,41,42,43,44,45,46,47]
Optimisation Methodology Development	Studies introducing or improving optimisation algorithms or hybrid optimisation techniques	[7,29,48,49,50,51,52]
Optimisation Model Development	Studies developing surrogate or machine-learning models for building performance prediction within optimisation workflows	[5,8,30,53,54]
Performance Assessment	Studies evaluating optimisation results or benchmarking performance of optimisation algorithms	[32,33,55,56,57]
Decision-Support Tool Development	Studies translating optimisation results into tools supporting designer decision-making	[34,35]

Table 3. Classification of reviewed studies according to the research concept.

Category	Description	References
Optimisation Framework Development	Studies proposing structured optimisation workflows integrating modelling, optimisation, and evaluation	[1,2,3,6,8,9,10,11,28,31,36,37,38,39,40,41,42,43,44,45,46,47]
Optimisation Methodology Development	Studies introducing or improving optimisation algorithms or hybrid optimisation techniques	[7,29,48,49,50,51,52]
Optimisation Model Development	Studies developing surrogate or machine-learning models for building performance prediction within optimisation workflows	[5,8,30,53,54]
Performance Assessment	Studies evaluating optimisation results or benchmarking performance of optimisation algorithms	[32,33,55,56,57]
Decision-Support Tool Development	Studies translating optimisation results into tools supporting designer decision-making	[34,35]

3.2. Optimisation Problem Characterisation

Zeng et al. [11] found optimisation problem formulation as challenging, requiring a comprehensive understanding prior to effective implementation and solution development. Primarily optimisation problems can be classified based on the characteristics, including number of objectives, nature of functions, constraints, type of variables, landscape and determinacy [11]. Depending on the number of objectives pursued, optimisation problems are classified as single-objective optimisation (only one objective function is optimised), multi-objective optimisation (two or three objective functions are optimised) and many-objective optimisation (more than three objectives) [11]. Function form of the optimisation can be either linear or non-linear and depending on that, the optimisation problem will be affected. If the optimisation problem has no constraints in the process of carrying out optimisation it is called an unconstrained optimisation problem, but the majority of research studies have defined a set of constraints called constrained optimisation [31]. Optimisation problem can be further classified depending on the type of design variable. Design variables can be either discrete design variables, continuous design variables and mixed integer design variables [5]. When the objective function has one peak, it is called unimodal. Moreover, optimisation problems can be defined as deterministic if the design variables are subject to small uncertainty or no uncertainty. In contrast, when the design variables are subject to uncertainty, it is called probabilistic uncertainty. The above-discussed characteristics of optimisation problems are evidence that primarily building optimisation problems greatly vary with objective function structure and variable type, ultimately influencing the selection of optimisation strategy as presented in Figure 2.

3.3. Building Function

An assessment of building typologies adopted in cost- and carbon-oriented building design optimisation studies is essential to understand how optimisation objectives have been aligned with building function in prior research [5]. As shown in Figure 3, the reviewed literature is dominated by residential buildings, with seventeen studies focusing on this typology. These studies primarily address thermal comfort, daylighting performance, energy efficiency, carbon emissions, and cost optimisation. The prominence of residential buildings in optimisation research can be attributed to their standardised design characteristics, relatively simpler modelling requirements, and strong cost and energy performance considerations, which make them suitable for parametric modelling and optimisation analyses. Notably, most residential case studies involve mid-rise developments, including terrace housing, apartment blocks, and multi-family residential buildings [31,34].

In comparison, nine studies investigating office buildings concentrate mainly on OC emissions, building energy performance, operational cost assessment, and structural safety. Studies focusing on educational (5) and other public buildings (6) predominantly emphasise EC reduction and life-cycle cost optimisation, often at the level of building envelopes or structural systems [2].

From a building function perspective, a clear research gap is evident. Despite escalating housing costs and the significant contribution of residential buildings to overall carbon emissions, optimisation studies addressing single and two-storey standalone houses remain limited. This is particularly significant given that low-rise residential buildings exhibit greater design variability, context sensitivity, and construction diversity compared to standardised mid-rise typologies. Consequently, the relative scarcity of optimisation research focusing on standalone residential houses highlights a critical knowledge gap and underscores the need for targeted BDO frameworks that explicitly address this building typology.

3.4. Building Design Variables

The selection of appropriate design variables constitutes a critical phase in BDO [9] as it defines the feasible design space and directly influences optimisation outcomes. Yang et al. [58] revealed that optimisation modelling should begin with the careful identification of decision variables, as poorly formulated variable sets can restrict design diversity and lead to suboptimal solutions. Hence, the identification of variables that meaningfully affect the selected objective functions is essential for developing robust optimisation frameworks [9].

Zeng et al. [11] classified design variables into design parameters and operational parameters, reflecting the interdependence between design-stage decisions and operational performance. Extending that classification, design parameters may be continuous or discrete [47], while operational parameters may be continuous, multi-step, or binary, depending on system behaviour and control logic.

After careful evaluation of the design variables employed in the reviewed studies, a total of 88 design variables were identified and visually presented in Figure 4 as a Sankey diagram. The Sankey diagram was developed based on a structured coding protocol designed to systematically identify, classify, and quantify building design variables reported in the reviewed studies. The coding process followed three predefined rules. First, only variables directly incorporated into optimisation studies were considered. In that phase, each article was examined to extract design variables explicitly employed as optimisation parameters within their study. Variables that were discussed in the literature section but not directly incorporated into the optimisation process were excluded from the analysis.

Second, each variable was counted once per study, regardless of the number of times it appeared within the article, thereby preventing over-representation of variables discussed extensively in individual studies. Third, to ensure consistency and avoid duplication during frequency analysis, variables representing the same design concept but reported using different terminology were standardised under a unified coding scheme. For example, “WWR”, “window-to-wall ratio”, “glazing ratio”, and “window area-to-wall area ratio” were coded as WWR, while “orientation angle”, “building orientation”, and “azimuth angle” were classified as building orientation. This standardisation process ensured that conceptually equivalent variables were consistently classified across all reviewed studies.

The extracted variables were subsequently categorised into four main groups according to their functional role within the building design process: (i) geometry and massing variables, (ii) building envelope variables, (iii) building structure variables, and (iv) building systems and operational variables.

To enhance coding reliability, the extracted variables, coding classifications, and frequency counts were independently reviewed by three researchers. Any discrepancies in variable interpretation, classification, or category allocation were discussed and resolved through consensus prior to the final frequency analysis. The frequency of each variable category and sub-variable was subsequently aggregated across all reviewed studies and visualised using the Sankey diagram to illustrate the hierarchical structure and relative prominence of building design variables within the BDO literature.

The building envelope cluster is the most dominant, with 112 repetitions, indicating a strong research focus on envelope-driven optimisation. Wall, window, roof, floor, and shading variables collectively influence both material intensity and thermal performance, making the envelope a critical contributor to embodied and operational impacts and capital cost. Wall-related variables, particularly insulation thickness, insulation material type, and wall material selection, directly increase EC and cost through additional material layers and the use of carbon-intensive products. Window-related variables show particularly high repetition, with window-to-wall ratio (WWR), window U-value, and SHGC emerging as dominant parameters. Roof and floor variables, especially insulation thickness and U-values, further contribute to embodied impacts. The dominance of this cluster reflects the envelope’s role as a high-leverage design domain for balancing EC, capital cost, and operational performance.

The building system and operational cluster includes 36 repetitions, encompassing HVAC parameters, lighting loads, ventilation variables, and renewable energy systems. While the above cluster is primarily associated with operational energy, operational cost and capital cost are also affected through equipment sizing and system selection.

Geometry and massing variables account for 32 repetitions, highlighting the importance of early-stage form-related decisions. This cluster is dominated by building orientation, followed by total floor area, building length and width, and aspect ratio. Building geometry variables strongly influence material quantities by determining building compactness, envelope surface area, and structural spans [37]. Inefficient geometry typically increases the volume of structural and envelope materials, leading to higher EC and CCs. As such, geometry and massing variables act as upstream drivers that establish the baseline for EC and capital cost outcomes before envelope and structural decisions are made.

Building structure variables account for 26 repetitions and include member dimensions, reinforcement ratios, material grades, and structural system characteristics. Although less frequently optimised, structural variables exert a disproportionately large influence on upfront EC and capital cost, as structural materials typically dominate both emissions and expenditure. Beam and column dimensions and reinforcement ratios directly determine concrete and steel quantities, while material grades influence emission intensity and unit costs.

3.5. Objective Functions

Determination of objective functions represents the core of optimisation problem formulation in BDO [6]. As discussed in Section 3.2, optimisation studies may be classified as single-objective, multi-objective, or many-objective, depending on the number and interaction of objective functions considered. Once objective functions are defined, the selection of appropriate optimisation parameters and design variables follows accordingly [11]. Given that this study is primarily driven by low-carbon and cost-effective building design, a comprehensive evaluation of objective functions and their associated performance parameters was undertaken across the reviewed studies.

Based on this analysis, nine primary objective function categories were identified: energy performance, daylighting performance, thermal comfort, EC emissions, OC emissions, construction and material cost, operational cost, indoor air quality, and structural safety. As the search strategy deliberately centred on carbon emissions and cost, it is expected that variants of these objectives dominate the dataset. Consistent with this expectation, EC emissions and construction and material cost objectives together account for approximately 38% of the reviewed studies, while operational carbon emissions and operational cost objectives collectively account for approximately 32%. Consequently, nearly 70% of the reviewed optimisation studies prioritise cost- and carbon-related objectives, confirming the continued dominance of a cost–carbon efficiency paradigm in optimisation research.

A critical gap identified in this review relates to the inconsistent specification and implementation of life-cycle boundaries across the analysed studies. In accordance with BS EN 15978 [59], building life-cycle assessment should systematically encompass the product stage (A1–A3), construction process stage (A4–A5), use stage (B1–B7), and end-of-life stage (C1–C4), with Module D reported separately. However, the findings reveal substantial heterogeneity in how these boundaries are operationalised in optimisation studies. While many studies nominally refer to “life cycle” cost or carbon, the actual system boundaries are often partial and fragmented, ranging from product-stage-only assessments (A1–A3) to operational-only evaluations (B6), with only a limited subset adopting comprehensive A1–C4 coverage. A dominant trend is the over-reliance on operational energy (B6) as a surrogate for whole-life performance, with upstream processes (A4–A5) and end-of-life stages (C1–C4) frequently omitted [5,9,37,39,41]. Similarly, “life-cycle cost” is commonly reduced to initial construction cost (A1–A5 proxy) and operational energy expenditure (B6), neglecting maintenance (B2), replacement (B4), and disposal-related costs [39,49,50]. This methodological fragmentation reflects divergent modelling assumptions and data constraints but ultimately results in inconsistent objective definitions and reduced comparability across studies. Consequently, reported cost–carbon trade-offs may be systematically biased due to truncated system boundaries. This inconsistency in the system boundary definitions for the life cycle stage (EN 15978) directly influences the reported optimisation outcomes. Studies employing truncated boundaries tend to underestimate total carbon or cost impacts, depending on the excluded life-cycle stages. For instance, operational-focused assessments may significantly undervalue embodied carbon contributions, while product-stage-only analyses fail to capture use-phase and end-of-life effects. Consequently, the magnitude and direction of cost–carbon trade-offs can vary substantially across studies, leading to potentially biased optimisation results. This variability reduces the comparability of findings and may result in misleading conclusions regarding optimal design strategies. These findings underscore the necessity for standardised and explicitly reported life-cycle boundaries aligned with EN 15978 [59], particularly in early-stage BDO, to improve methodological transparency and enable robust cross-study synthesis.

Objective-Driven Mapping of Optimisation Parameters and Building Design Variables

Following the analysis of objective functions, a systematic cross-mapping between optimisation objectives and the building design variables most frequently employed in the reviewed studies was conducted, as presented in

Table 4. Objective-driven mapping of optimisation parameters and building design variables.

Objective Function	Optimised Parameters	Design Variables
Energy Performance (heating energy demand, cooling energy demand)	EUI (Energy Use Intensity)	Geometry & massing Envelope thermal properties Fenestration & glazing Shading & solar control HVAC systems & controls Lighting & internal loads Airtightness & ventilation Renewable energy systems
Daylighting	UDI (Useful Daylight Illuminance) SDA (Spatial Daylight Autonomy) Illuminance	Geometry & massing Fenestration & glazing Shading & solar control Lighting control systems
Thermal Comfort	PMV (Predicted Mean Vote) PPD (Predicted Percentage of Dissatisfied) DH (Discomfort Hours) Operative Temperature and RH (Relative Humidity)	Envelope thermal performance Fenestration & shading HVAC systems & controls Geometry & massing Airtightness & ventilation
Indoor Air Quality	Indoor CO2 concentration	Ventilation systems Fenestration operability Envelope airtightness HVAC system
Carbon Emission	Embodied Carbon (kgCO2e)	Structural system & sizing Geometry & massing Building Envelope size and materials
	Operational Carbon (kgCO2e)	Envelope thermal performance Fenestration & shading HVAC systems Renewable energy systems Airtightness & ventilation
Cost	Material Cost Investment Cost Construction Cost	Geometry & massing Structural design Material selection Envelope systems
	Global Cost Energy Cost	Envelope thermal performance HVAC systems & controls Lighting systems Renewable energy systems
Structural Safety	Weight Displacement Capacity Ratio	Structural system configuration Structural member sizing Material grades Imposed loads

. The objective function variable mapping was developed through a qualitative inductive synthesis of the reviewed studies. Rather than applying a quantitative co-occurrence analysis, the mapping was derived by systematically examining the optimisation objectives and the corresponding design variables used in each reviewed study. In particular, Table 4 should be interpreted as a qualitative decision-support guide rather than a statistically derived representation of optimisation objective-design variable relationships.

This cross-mapping explicitly demonstrates that the selection of design variables is inherently objective-dependent, with different objective functions activating distinct subsets of design variables across geometry, envelope, systems, materials, and structural domains. Rather than treating design variables as a fixed or exhaustive set, the proposed mapping highlights how objective functions implicitly define the relevant design space and influence optimisation problem formulation. Accordingly, Table 4 serves as a practical decision-support guide, enabling researchers and practitioners to identify and select appropriate building design variables aligned with their chosen optimisation objectives.

3.6. Constraints

Constraints define the limitations that must be satisfied to ensure acceptable, feasible, and practically implementable solutions in BDO studies [30]. In the reviewed literature, constraints are applied to both objective functions and design variables, thereby restricting the feasible design space and guiding the optimisation process toward realistic outcomes [8]. Accordingly, ref. [9] emphasised that in the absence of appropriately formulated constraints, optimisation algorithms may converge to solutions that are technically infeasible or practically unacceptable in MOO problems.

These constraints can be classified in several complementary ways. Primarily, based on their degree of strictness, constraints may be categorised as hard or soft. Hard constraints are typically expressed as explicit equations or inequalities and must be satisfied without any violation [60]. Designs violating these constraints are treated as infeasible and excluded from the solution space. In contrast, soft constraints allow limited violations, which are accounted for by introducing penalty terms into the objective function [61]. This transformation converts a constrained problem into an unconstrained or weakly constrained one, enabling more efficient solutions by certain optimisation algorithms. Soft constraints are particularly useful in BDO, where performance requirements such as thermal comfort do not need to be met perfectly at all times [11].

The analysis of constraint formulation across the reviewed studies reveals a pronounced imbalance between hard and soft constraints, with important implications for optimisation robustness and solution realism. Hard constraints are predominantly economic and technical, such as budget limits [30,37], NPV and payback thresholds [9], and energy balance requirements [8,55], which strictly delimit the feasible solution space. In contrast, occupant-centric and environmental performance aspects are frequently treated as soft constraints, embedded within simulation outputs rather than explicitly constrained. For instance, thermal comfort [30,41,44,50] and daylighting metrics [1,28,52] are often implicitly controlled through objective functions or post-processing, allowing trade-offs that may compromise user well-being. Notably, some studies partially formalise these as hard constraints (e.g., DH ≤ threshold [37], comfort > 85% [39]), indicating emerging hybrid approaches.

3.7. Standards and Codes

Building design standards and codes play a decisive methodological role in shaping BDO, systematically embedding system boundaries, calculation procedure and constraints. Over various standards and codes published for governing built environment practices, ISO standards and European (EN) standards are predominantly employed in carbon and cost-focused reviewed studies. In contrast, ASHRAE standards are widely employed in energy performance and thermal comfort-oriented research, including the determination of room air change or penetration rates [8], the evaluation of total energy consumption [56], and the definition of acceptable ranges for indoor operative temperature [50].

Carbon-related optimisation studies are predominantly structured around ISO 14040/44 [62] and EN 15978 [59]. These standards define LCA principles, reporting stages, and modular boundaries, enabling the explicit integration of EC and OC within MOO frameworks [8,30]. Ref. [39] adopted EN 15804 [63], the European standard for Environmental Product Declarations (EPDs), as the primary source for material-specific emission factors, thereby defining the system boundary for EC emissions associated with the materials included in the database.

In contrast, cost optimisation-focused studies remain comparatively fragmented. Although standards such as EN 15459-1 [8], and ISO 15686-5 [35] provide structured methodologies for LCC assessment, their adoption is less widespread than carbon standards. Cost optimisation is further constrained by the absence of universally accepted CC databases, forcing researchers to rely on regional price books, historical cost data, or national economic indices. This variability introduces uncertainty and limits the reproducibility of cost-focused optimisation outcomes.

A critical distinction emerges between carbon and cost domains: while carbon optimisation benefits from globally harmonised standards and EPD-based datasets, cost optimisation is inherently context-dependent and data-sensitive. As a result, carbon objectives are more frequently prioritised or weighted more robustly in MOO models, while cost objectives are often treated as secondary or scenario dependent.

In addition to the above standards, energy performance regulations and policy-driven frameworks also influence the formulation of optimisation problems in several studies. Accordingly, Ascione et al. [37] anchored the developed optimisation framework to the Energy Performance of Buildings Directive (EPBD) in the European Union and to the Nearly Zero Energy Buildings (NZEB) standard, which establishes regulatory targets for building energy consumption and carbon emissions. These frameworks are frequently incorporated as performance constraints or benchmark targets within optimisation models, guiding the exploration of design solutions that meet mandated energy efficiency requirements while balancing other objectives such as carbon emissions and cost [6].

3.8. Optimisation

3.8.1. Optimisation Strategies

Optimisation constitutes the core of BDO, with the selection of an appropriate optimisation strategy determining how the optimisation process is formulated and executed [36]. A review of the existing literature indicates that two principal optimisation strategies are predominantly adopted: single-algorithm optimisation, in which a standalone optimisation algorithm is directly coupled with performance simulation models, and surrogate-assisted optimisation, where computationally efficient surrogate models are integrated with optimisation algorithms to alleviate simulation burden while preserving solution fidelity.

To improve the transparency of the algorithm analysis, the reviewed studies were classified using a four-layer hierarchical structure (

Table 5. Distribution of optimisation strategies.

Layer 1: Optimisation Strategy	Layer 2: Surrogate Model	Layer 3: Optimisation Algorithm	Layer 4: Algorithm Family	No. of Studies	Average No. of Objectives	Average No. of Design Variables	References
Single-algorithm optimisation	–	NSGA-II	EM	15	2.8	9.5	[3,9,28,30,33,34,35,41,44,45,48,49,55,56,57]
	–	NSGA-III	EM	1	3	9	[41]
	–	MOGA	EM	3	3	15	[6,37,51]
	–	GA	EM	2	2	10	[1,53]
	–	MOQGA	EM	1	6	11	[5]
	–	MOPSO	SBM	1	3	11	[55]
	–	Harmony Search	EM	1	2	6	[47]
	–	HypE	EM	1	2	4	[39]
	–	SPEA2	EM	1	7	15	[39]
	–	PSO	SBM	1	3	9	[41]
Surrogate-assisted optimisation	ANN/BPNN	NSGA-II	EM	4	2.5	16.3	[10,40,44,50]
	ANN/BPNN	GA	EM	1	3	13	[2]
	DNN	NSGA-II	EM	1	2	9	[42]
	GAN	NSGA-III	EM	1	2	44	[8]
	Random Forest	NSGA-II	EM	1	3	8	[36]
	RF + GB	GA, RBFOpt, CMA-ES, NSGA-II, RBFMOpt	EM	1	4	10	[31]
	SVM	NSGA-II/III	EM	1	3	12	[41]
	ELM/ENR	NSGA-II	EM	1	3	8	[46]
	Kriging	NSGA-II	EM	1	3	12	[43]
	Kriging	NSGA-II, NSGA-III, PESA-II, SPEA2	EM	1	3	29	[29]

) comprising (i) optimisation strategy, (ii) surrogate model, (iii) optimisation algorithm and (iv) optimisation algorithm family. The first layer distinguishes between single-algorithm optimisation and surrogate-assisted optimisation approaches. The second layer identifies the surrogate model employed, where applicable, while the third layer records the optimisation algorithm used within each study. To account for differences in optimisation problem complexity, the analysis incorporates two additional indicators: average number of optimisation objectives and average number of design variables associated with each algorithm. The average number of objectives was calculated by averaging the total number of objective functions reported across all studies employing a particular algorithm, while the average number of design variables was obtained by averaging the total number of decision variables considered in those studies. These indicators were included as proxy measures of optimisation problem complexity, where a larger number of objectives reflects increased complexity in the optimisation formulation and a larger number of design variables represents a higher-dimensional design space. Consequently, algorithms associated with higher average values can be interpreted as being applied to comparatively more complex optimisation problems.

Furthermore, explicit statistical rules were established for studies employing multiple optimisation algorithms. When more than one optimisation algorithm was evaluated within a single study, each algorithm was recorded separately in the frequency analysis to accurately represent its utilisation. However, because all algorithms within a given study were applied to the same optimisation problem formulation, the corresponding number of optimisation objectives and design variables remained identical for each algorithm.

• Single Algorithm optimisation Strategy

The reviewed studies exhibit a strong reliance on metaheuristic optimisation algorithms which are based on stochastic population [3], reflecting the complex, non-linear, and multi-objective nature of BDO problems involving cost, carbon emissions, energy performance, and comfort trade-offs. Evolutionary metaheuristic (EM) algorithms dominate the literature as presented in Table 5. Within this category, NSGA-II developed in 2002 is the most extensively adopted algorithm, representing over 50% of total applications [8]. Its widespread adoption is attributed to its ability to efficiently generate well-distributed Pareto-optimal solution sets while preserving population diversity through non-dominated sorting, crowding-distance-based selection, and elitist retention mechanisms [10,31]. These features enable effective exploration of high-dimensional, non-linear, simulation-driven design spaces, making NSGA-II particularly appropriate for early-stage building design optimisation [9].

Conventional Genetic Algorithms (GAs) appear less frequently and are primarily applied in simplified optimisation formulations, due to their limited local search capability [36]. Advanced evolutionary algorithms including NSGA-III and HypE are limitedly employed, suggesting that their higher implementation complexity and computational requirements may limit widespread adoption in building optimisation studies [8].

Swarm-based metaheuristics (SBMs), including PSO/MOPSO, constitute a smaller proportion of applications. Although these algorithms are recognised for their rapid initial convergence and relatively low computational overhead, comparative studies indicate that MOPSO demonstrates slower convergence towards optimal Pareto solutions than NSGA-II in complex BDO problems [30]. Moreover, Liu et al. [36] reported that PSO algorithms demonstrate limited late-stage convergence improvement and reduced solution diversity, constraining their effectiveness in highly MOO contexts. Human-based algorithms, such as Harmony Search, are rarely adopted, indicating limited penetration beyond niche structural optimisation applications [47].

Overall, the findings highlight a pronounced methodological concentration around NSGA-II, underscoring its suitability for simulation-based BDO.

2.

Surrogate-Assisted Optimisation Strategy

Around 34% of studies adopted the surrogate-assisted optimisation (SAO) approach, primarily driven by the prohibitive computational cost of high-fidelity simulation engines such as EnergyPlus, TRNSYS, Radiance, and BIM-integrated performance tools. Direct coupling of evolutionary algorithms with such simulation environments often results in excessive runtimes, particularly for multi-objective problems [42,64]. In this context, surrogate models play a crucial role in accelerating objective function evaluations and enabling efficient exploration of complex design spaces [46].

Among surrogate techniques, Artificial Neural Networks (ANNs) dominate the reviewed literature due to their strong capacity to approximate highly non-linear relationships between design variables and performance objectives [31,50]. Rizehbandi et al. [31] reported that ANN-based models are proficient in carbon and energy prediction, particularly when large and representative datasets are available. However, their effectiveness is contingent upon the availability of sufficient training data, which poses a significant limitation during early conceptual design stages where sampling budgets are typically constrained [34].

Ensemble learning methods such as Random Forests (RFs) and Gradient Boosting (GB) appeared in BIM-enabled optimisation contexts, where robustness to noisy data and moderate interpretability are desirable [31]. GB and RFs have shown superior predictive accuracy for CO₂ emissions and energy consumption compared to single learners, reinforcing their suitability for data-driven optimisation workflows [36].

More advanced surrogate strategies including Kriging (Gaussian Process models), Support Vector Machines (SVMs), and Extreme Learning Machines (ELMs) are employed in structural and life-cycle optimisation problems, where controlled design-of-experiment settings and smoother response surfaces are required [29,43]. Emerging approaches, including GAN-based data enhancement remain limited in number but show promise in addressing data scarcity and accelerating convergence in the complex optimisation process [8].

Across SAO studies, Latin Hypercube Sampling (LHS) is the predominant sampling technique, favoured for its space-filling properties and suitability for high-dimensional parametric studies [8,10]. Alternative approaches such as Sobol sequences [41] and Monte Carlo sampling is adopted in cases where uncertainty quantification or reduced experimental runs are required. According to the reviewed findings, surrogate performance is validated using cross-validation and statistical accuracy metrics, including R², RMSE, MAE, and MAPE, ensuring reliability prior to optimisation integration [36,43].

3.8.2. Optimisation Results Evaluation and Validation

Results Evaluation

All the reviewed papers presented their optimisation outputs as a Pareto front, the set of best solutions, where improving one objective means sacrificing another. Therefore, evaluation of Pareto solution sets and the derivation of informed decisions involve multiple analytical phases. Based on the synthesis of post-optimisation evaluation practices reported in the reviewed literature, seven distinct methods are identified for the post-optimisation phase, which together provide a structured methodological flow for transforming optimisation results into meaningful and decision-relevant study outcomes.

• Pareto front generation and trade-off exploration: Depending on the number of objective functions considered, Pareto fronts are visualised in two- or three-dimensional objective space to provide an initial understanding of the trade-off structure between competing objectives. This stage commonly employs visual Pareto-front inspection, formal Pareto dominance definitions, and non-dominated sorting with crowding distance to identify the Pareto set and qualitatively examine objective conflicts.
• Pareto-set quality, convergence, and robustness assessment: To ensure that the identified Pareto front is reliable and well-converged, several studies evaluate solution quality using hypervolume indicators [10,29,34] and multi-run robustness and stability checks [6,28]. These analyses verify convergence behaviour, diversity, and consistency of the solutions, thereby preventing premature interpretation of incomplete or unstable fronts.
• Extreme-point identification and baseline comparison: At this stage, extreme or anchor solutions such as minimum-carbon or minimum-cost designs are extracted from the Pareto set to define performance bounds and support objective-specific decision scenarios. Hong et al. [30], Kang et al. [41] and Lin et al. [49] further compared these extreme points against baseline or reference designs to quantify achievable performance improvements and contextualise optimisation benefits.
• Balanced or compromise solution identification: To identify efficient trade-off solutions within the Pareto set, ref. [11] adopted distance-to-ideal (utopia-point) methods, whereby objectives are normalised and the Pareto solution with the minimum distance to the ideal point is selected. Alternatively, adopting a fitness function, the Pareto front solution in its original order is shortened [1,30,36].
• Design-variable influence assessment: To improve the interpretability of optimisation outcomes, sensitivity and correlation analyses are frequently conducted on Pareto solutions to identify influential design variables and sensitive parameters [1,10,37]. This stage provides insight into the underlying drivers of optimal performance and supports knowledge extraction beyond numerical optimisation results.
• Solution-space reduction: When Pareto sets are large or high-dimensional, solution-space reduction techniques are applied to improve interpretability. Li et al. [65] and Shi et al. [28] employed K-means clustering to group non-dominated solutions and select representative alternatives closest to cluster centroids using Euclidean distance. Adopting the elbow criterion, supported by reductions in the sum of squared errors and peak silhouette coefficients, the number of clusters in both studies was evaluated. While this unsupervised approach improves interpretability and enables identification of representative design archetypes, its effectiveness remains sensitive to the predefined cluster number and distance metric, potentially overlooking extreme or stakeholder-preferred solutions.
• Decision support and stakeholder-oriented selection: Selection of a preferred design alternative is carried out under stakeholder preferences and practical constraints. This stage commonly employs multi-criteria decision-making (MCDM) techniques, including TOPSIS [6,10], equal-weighted or modified TOPSIS with correlation-based weighting [2], stakeholder-defined weighting schemes, and Weighted Sum Method [56] to rank Pareto solutions and identify final design candidates.

Results Validation

Validation is a critical component of BDO research, as Pareto-optimal solutions are only meaningful if the underlying models, optimisation frameworks, and resulting design configurations are reliable and practically feasible. As presented in

Table 6. Optimisation results validation approaches.

Validation Approach		References
Case Study	Single case study	[1,2,3,5,8,9,10,28,29,30,31,33,35,36,37,38,39,40,42,43,46,47,48,49,50,51,52,54,55,57]
	Multiple case study	[6,7,11,32,34,41,45,56]
	Hypothetical case study	[53]
Literature comparison		[33,34,52]
Building codes compliance		[44,46,48]

, all reviewed studies incorporated different forms of validation. The validation approaches can be broadly categorised into three main groups: (i) case study-based validation, including single, multiple, and hypothetical case studies; (ii) comparison of optimisation outcomes with findings reported in the previous literature; and (iii) verification against relevant building codes and standards. These approaches were adopted to establish the credibility of the proposed optimisation frameworks, evaluate the reliability of the generated Pareto-optimal solutions, and assess the practical applicability of the findings within realistic building design scenarios.

The dominant validation strategy is the case-study-based application, where optimisation methods are tested on real building designs. Samarasinghalage et al. [9] validated a MOO framework for BIPV through three application scenarios (canopy, roof sheets, and cladding), demonstrating both feasible Pareto-optimal solutions and constraint-driven infeasibility under strict economic limits. Similarly, an optimal decision support model by [5] and an optimisation framework developed by Kang et al. [41] validated their model and framework through utilising a case study of a public building and a zero-carbon building, respectively. In structural design optimisation, validation often relies on comparative case studies, demonstrating reductions in EC and cost relative to reference designs [45]. The primary reason for the dominance of case-study-based validation is that optimisation frameworks must be evaluated within realistic design contexts where building performance is influenced by factors such as climate conditions, building geometry, material properties, and regulatory constraints. In particular, simulation-based optimisation tools require detailed input data, which are most accurately represented through real building cases. In this sense, case studies demonstrate that theoretically developed optimisation frameworks and models are practically feasible, thereby ensuring the practical relevance of the research. Moreover, case studies provide a reliable approach for validating the robustness and real-world applicability of optimisation frameworks.

Another validation approach identified in the reviewed studies is the use of hypothetical case studies to evaluate the effectiveness of proposed optimisation models and frameworks. For example, Elsayed et al. [53] employed hypothetical building layout scenarios to demonstrate and assess the capability of their multi-objective optimisation framework. Rather than applying the model to an existing real-world building project, the authors developed controlled design scenarios that enabled systematic testing of the framework’s ability to generate optimal solutions while satisfying predefined design constraints. This approach provided an effective means of verifying the functionality, applicability, and performance of the proposed optimisation model during its development stage.

Code compliance and benchmark-based validation further strengthen reliability claims. Some studies validate optimisation outputs against recognised standards such as ASHRAE Standard 140 BESTEST [48], while others embed structural code requirements directly within the optimisation process to ensure the feasibility of all Pareto solutions [46]. These approaches provide additional assurance regarding the applicability and validity of optimisation results beyond individual case study contexts.

Validation through comparison with the prior literature is used to confirm the consistency of optimal design parameters with previously reported results [33,52]. Similarly, [34] evaluated the performance of the optimisation tool developed in their study against the research findings of three conceptual problems (office buildings) from the literature.

3.9. Tools and Software

BDO relies on a combination of software, plugins, libraries, and tools deployed across multiple stages. Consequently, no single platform supports the entire optimisation process; instead, studies adopt interoperable toolchains tailored to specific analytical tasks.

For geometric and parametric model development, widely adopted platforms include SketchUp, Autodesk Revit, Rhinoceros, Grasshopper, and OpenStudio. Parametric design environments are particularly prevalent in early-stage optimisation studies, as they preserve dynamic relationships between design variables and building geometry. Prior research consistently reports extensive use of Rhinoceros–Grasshopper, Dynamo–Revit, and SketchUp-based workflows during conceptual design stages. Among these, Grasshopper dominates early-stage optimisation research due to its visual programming interface and strong compatibility with environmental analysis plugins and optimisation engines [1]. In parallel, BIM-based parametric environments have gained traction, with tools such as Revit increasingly incorporating embedded energy simulation and life-cycle assessment capabilities through specialised plugins [36].

Simulation engines constitute the core performance-evaluation layer of optimisation workflows. EnergyPlus is the most widely applied simulation engine, followed by DesignBuilder, Radiance, and OpenStudio, to evaluate operational energy consumption, thermal comfort, daylight performance, and indoor environmental quality [65]. Within Grasshopper-based workflows, Ladybug and Honeybee function as key interfaces for climate analysis, energy simulation, daylight modelling, and thermal comfort assessment [10].

Optimisation engines play a central role in exploring high-dimensional design spaces and generating Pareto-optimal solution sets. Evolutionary multi-objective algorithms, particularly NSGA-II, dominate across platforms, including jEPlus, MATLAB, Python-based libraries (pymoo, PlatEMO) [9], Octopus, Wallacei, and MOBO [6]. Grasshopper-based optimisers enable rapid Pareto-front visualisation, whereas MATLAB and Python environments are preferred for algorithm customisation and surrogate-assisted optimisation [1].

3.10. Critical Dimensions Affecting Low-Carbon and Cost-Effective Building Design Optimisation

BDO is inherently computationally intensive and interdisciplinary, and its effectiveness depends not only on algorithm selection but also on robustness, accessibility, and practical integration within real design workflows [6]. The reviewed studies revealed that limitations arise across several critical dimensions, each of which directly influences the reliability and practical relevance of optimisation results. Challenges were categorised based on their primary point of emergence and functional role within the BDO workflow, enabling a process-oriented synthesis that distinguishes computational, modelling, data-related, decision-making, and socio-technical limitations.

Computational and algorithmic robustness is fundamental for effectively exploring large, high-dimensional design spaces and achieving well-converged Pareto fronts [41]. However, simulation-driven optimisation often suffers from excessive computational cost, scalability limitations, premature convergence, and insufficient convergence assessment, which could restrict solution diversity and compromise optimality [38].

Equally important is problem formulation and modelling, as optimisation performance depends on clearly defined objectives, constraints, and performance metrics [55]. Conflicting objectives, heterogeneous performance indicators, difficulties in enforcing regulatory constraints, and limited life-cycle scope reduce the robustness of results and hinder cross-study comparability, leading to ambiguous interpretations [31].

Reliable optimisation outcomes depend heavily on the quality, transparency, and representativeness of input data and assumptions [53]. Limited localisation of cost and EC datasets, uncertainty in early-stage assumptions, and sensitivity to economic and environmental parameters weaken transparency and confidence in results [36]. These issues are compounded by limited validation, incomplete reporting, and poor reproducibility.

The ultimate value of optimisation lies in its ability to support informed design decisions and guide practical implementation [45]. However, numerous studies report difficulties in translating complex Pareto fronts into actionable design choices, incorporating stakeholder preferences, and ensuring the constructability of optimised solutions [31]. Limited reusability of developed frameworks and case-specific findings further constrain the broader applicability of optimisation research. In addition, early design decisions often lead to carbon and cost lock-in, reducing the effectiveness of optimisation when applied at later stages.

Beyond technical considerations, the effectiveness of BDO is influenced by institutional, regulatory, and socio-technical factors. Barriers to industry adoption, misalignment with policy and regulatory frameworks, resistance to integrated design practices, and reliance on specific software toolchains limit interoperability and bias research outcomes [52].

Accordingly, drawing on the preceding analysis, key challenges in low-carbon and cost-effective BDO are presented in

Table 7. Key challenges and limiting factors in sustainable building design optimisation.

Category	Challenge	References
Computational & Algorithmic Dimension	High computational cost of simulation-driven optimisation	[3,8,10,11,30,37,38]
	Risk of local optima in GA-based methods	[5,10,41]
	Large and mixed-variable design spaces	[37]
	Algorithm scalability limitations	[10]
	Convergence uncertainty	[10,57]
Problem Formulation & Modelling Dimension	Conflicting objectives (energy–carbon–cost–comfort)	[2,3,5,8,36]
	Constraint handling and code compliance	[40]
	Weak or non-intuitive objective correlations	[2]
	Limited lifecycle boundaries (operational-only focus)	[8]
	Difficulty in integrating professional tools (LCA, cost, simulation)	[35,57]
	High expertise requirements for simulation and optimisation model setup	[52]
	Objective and performance-metric inconsistencies	[31]
Data & Knowledge Dimension	Limited localisation of embodied-carbon data	[53]
	Limited availability of data	[8,44,49]
	Uncertainty of the quality of data	[8,44,49]
	Early-stage uncertainty in assumptions related to model performance	[31,40]
	Uncertainty in tariffs, emission factors, and service life	[36]
	Black-box nature of ML-based optimisation	[8,42]
	Deficiencies in software tool configurations	[31]
Decision-Making & Practice Integration Dimension	Difficulty selecting solutions from Pareto fronts	[8]
	Stakeholder preference articulation	[11,39]
	Constructability and practicality of optimised designs	[45,57]
	Limited reusability of findings or developed frameworks/models	[1,45]
	Tool integration and fragmented workflows	[3,57]
	Limited accessibility of optimisation workflows for non-technical stakeholders	[31]
Policy, Adoption & Socio-Technical Dimension	Lack of interest in industry adoption	[44]
	Socio economic uncertainty	[31]
	Resistance to integrated design practices	[52]
	Toolchain bias and software dependency	[34]
	Limited interoperability across software ecosystems	[44]

.

4. Discussion

4.1. Optimisation Algorithm Selection Criteria

As per the reviewed findings, although numerous BDO studies adopt specific optimisation algorithms and algorithm selection is typically based on a limited set of considerations, including computational efficiency, convergence speed, or implementation convenience, it was observed that the existing literature largely lacks a systematic discussion of the comprehensive criteria that should be evaluated before selecting an optimisation algorithm for a given problem. From Liu et al.’s [36] perspective, inadequate algorithm selection could compromise optimisation performance and undermine the reliability of outcomes. Similar gaps have been addressed in other engineering disciplines through the development of context-specific evaluation frameworks and assessment models that systematically identify and synthesise key decision factors to support methodological selection and evaluation processes [66,67]. Despite the increasing application of optimisation techniques in BDO, limited attention has been given to the development of systematic guidance for optimisation algorithm selection. Addressing the above issue, this study synthesises evidence from selected research articles to establish algorithm selection criteria (

Table 8. Optimisation algorithm selection criteria.

No	Selection Criteria	Brief Explanation	References
A1	Problem Complexity and Non-Linearity	Building simulation outputs are often non-linear, discontinuous, and multi-modal. Derivative-free evolutionary and heuristic algorithms are better suited to such problems, as they are robust to irregular search landscapes and capable of escaping local optima.	[37]
A2	Computational Cost and Efficiency	When optimisation is coupled with time-intensive simulation engines, algorithms must balance solution quality with computational efficiency. Metaheuristic algorithms generally outperform exhaustive or sequential search methods under constrained computational budgets.	[8,9,11,31,34,37,40,46,56]
A3	Ability to Maintain Pareto Solution Diversity	In MOO, generating a well-distributed Pareto front is essential for informed decision-making. Algorithms with explicit diversity-preservation mechanisms enable broader and more meaningful exploration of trade-off solutions.	[8,10,34,41]
A4	Scalability with Number of Design Parameters	Scalability describes how algorithm performance changes as the dimensionality of the design space increases. Evolutionary algorithms generally exhibit superior scalability and reliability compared to swarm-based or sequential methods in large-scale optimisation problems.	[1,9,36]
A5	Robustness and Convergence Performance	Robustness refers to an algorithm’s ability to consistently produce high-quality solutions across multiple runs, while convergence performance reflects the speed and reliability of approaching optimal solutions. These properties are commonly evaluated using indicators such as hypervolume.	[29,57]
A6	Availability of Data	Data availability influences algorithm selection, particularly for surrogate-assisted optimisation. When training data are limited, suitable optimisation algorithms are required to extract reliable solutions from sparse or incomplete datasets.	[48]
A7	Number of Optimisation Objectives	Optimisation algorithm performance is strongly influenced by whether a problem is single-objective, multi-objective, or many-objective. Multi- and many-objective problems require algorithms capable of generating Pareto-optimal solution sets to explicitly represent trade-offs among conflicting objectives.	Author synthesis
A8	Type of Design Variables	Building design problems involve continuous, discrete, and mixed-integer variables. Gradient-based methods struggle with non-differentiability and discontinuities, whereas evolutionary algorithms can robustly handle mixed variable types without requiring derivative information.	Author synthesis
A9	Compatibility with Simulation and Modelling Tools	Practical applicability depends on seamless integration with commonly used simulation and parametric modelling platforms. Algorithms that readily interface with tools such as EnergyPlus, GenOpt, and Rhino–Grasshopper enable efficient end-to-end optimisation workflows.	Author synthesis
A10	Flexibility and Ease of Implementation	Algorithms requiring fewer control parameters and less problem-specific tuning are easier to implement and adapt across diverse optimisation contexts, particularly in interdisciplinary design environments.	Author synthesis
A11	Potential for Hybrid Optimisation Approaches	Hybrid optimisation strategies that combine global and local search methods or integrate surrogate models can improve convergence accuracy and reduce computational burden in complex, simulation-based optimisation problems.	Author synthesis

), providing structured guidance for selecting appropriate optimisation algorithms in BDO problems.

The optimisation algorithm selection criteria presented in Table 8 were developed through the review of the selected studies. It was observed that while only a few studies explicitly stated the criteria used for optimisation algorithm selection, many others indirectly justified their choices through the evaluation and comparison of algorithm performance. Furthermore, during the review process, the authors identified several additional factors that should be considered when selecting an optimisation algorithm based on the characteristics of building design optimisation problems. Accordingly, Table 8 presents the proposed algorithm selection criteria derived from both the findings reported in the reviewed literature and the authors’ synthesis of key influencing factors identified across the studies. Therefore, the criteria represent a comprehensive literature-informed framework that incorporates both explicitly stated and implicitly inferred considerations for optimisation algorithm selection.

While these criteria provide a structured and literature-grounded basis for algorithm selection, their relative importance is likely to vary across project types, design stages, and practical contexts. Consequently, further refinement through engagement with industry practitioners and researchers actively involved in building optimisation is recommended to validate, prioritise, and rank these criteria according to their real-world significance. Such practitioner-informed weighting would enhance the applicability, robustness, and decision-support value of the proposed criteria.

4.2. ICCO-BD Framework: Integrated Carbon–Cost Optimisation Framework for Building Design

Analysis of the reviewed studies revealed several persistent limitations in the integrated consideration of cost and carbon objectives. Most cost-focused optimisation studies predominantly assess global or operational costs, while construction-phase costs, particularly those associated with building envelope components such as walls, roofs, and windows, remain underexplored. Similarly, carbon-oriented optimisation research largely concentrates on OC emissions, with comparatively limited attention given to EC impacts. Even among studies addressing EC, assessment is frequently restricted to cradle-to-gate stages (A1–A3), often excluding construction and transportation phases (A4–A5) due to data availability challenges and modelling complexity.

Furthermore, studies that evaluate CCs tend to focus primarily on structural systems, with limited integration of envelope-related cost components. Existing research is also skewed toward commercial and high-rise buildings, while optimisation studies targeting residential buildings remain scarce, largely due to difficulties in accessing reliable cost and carbon datasets. These gaps are particularly critical given that key design decisions made during the conceptual or early design stages exert a substantial influence on both CC and upfront EC emissions because once construction commences, cost and upfront EC impacts become largely locked in and difficult to mitigate.

To address the above limitations, this study develops an integrated conceptual framework derived from the reviewed literature findings to guide low-carbon and cost-effective BDO, particularly targeting upfront EC and CCs (Figure 5). By structuring the optimisation process into four interconnected phases, the framework establishes a cohesive end-to-end workflow that enhances methodological coherence while delivering measurable advantages, including improved time efficiency through reduced iterative redesign, enhanced solution quality via more comprehensive Pareto front exploration, increased consistency by minimising input–output mismatches, and stronger decision support through clearer interpretation of trade-offs for early-stage building design.

Phase 01—Optimisation Problem Formulation and Background Preparation

Phase 01, formulated through five stages, establishes the methodological foundation of the proposed framework and is central to ensuring the reliability and interpretability of optimisation outcomes. A recurring limitation in existing BDO studies is the emergence of sub-optimal or misleading solutions resulting from poorly articulated objectives, inconsistent constraint definitions, and ad hoc selection of design variables. Such shortcomings often lead to ill-posed optimisation problems and restrict the practical relevance of derived solutions.

In response, the proposed framework adopts a structured problem formulation approach grounded in the optimisation problem characteristics discussed in Section 3.2. Clear system boundaries are defined at the outset, followed by the explicit formulation of objective functions and associated constraints to ensure that performance targets remain realistic and design-stage appropriate. Particular emphasis is placed on aligning objective functions with early-stage decision-making to minimise CC and upfront EC.

Design space formulation (Stage 3) constitutes a central component of this phase. As cross-mapped in Table 4, design variables are systematically selected based on their demonstrated influence on CC and upfront EC, with primary focus on building structure and envelope-related parameters. Design variables are categorised as continuous, discrete, or mixed, reflecting both design flexibility and modelling feasibility. To maintain regulatory realism, variable bounds and feasibility constraints (soft constraints and hard constraints) are defined in accordance with applicable building codes and standards.

As the fourth stage, strategic selection of the optimisation paradigm is undertaken. Depending on the problem dimensionality, objective complexity, and computational burden, single-algorithm optimisation or surrogate-assisted optimisation strategies may be adopted. Algorithm selection is guided by clearly defined criteria (Section 4.1), ensuring compatibility with problem structure and convergence requirements. Population-based optimisation algorithms represent the dominant and widely accepted class of methods for addressing multi-objective problems characterised by conflicting objectives and non-linear design spaces in architectural optimisation. Moreover, their capability to generate well-distributed Pareto-optimal solution sets makes them particularly suitable for BDO. Accordingly, the subsequent optimisation phase is formulated based on population-based MOO algorithm selection.

In the 5th stage, a baseline building design is established to quantify reference values for EC, CCs, and selected performance metrics, including daylight performance and thermal comfort. The inclusion of daylight and thermal comfort assessments ensures that carbon- and cost-optimised design solutions remain operationally viable and do not compromise indoor environmental quality. This benchmark enables systematic performance comparison and strengthens the interpretive value of Pareto-optimal solutions.

Phase 02—Parametric Modelling and Performance Evaluation

Phase 02 translates the theoretically formulated optimisation problem into a computationally operable workflow. This phase comprises two stages. In Stage 6, a parametric or generative building model is developed using appropriate generative building design software, incorporating the design variables and constraints defined in Phase 01. Relevant carbon and cost datasets, together with simulation engines and assessment models, are then integrated to enable automated performance assessment. For data consistency and reliability, it is recommended that region- and project-specific cost datasets be utilised, while embodied carbon assessments should be based on Environmental Product Declarations (EPDs) developed in accordance with internationally recognised standards such as ISO 14025 [68] and EN 15804 [63]. Once integration is verified, in Stage 7, simulation-based evaluations are conducted for each generated design alternative. For every parametric iteration, upfront EC and CCs with either one or two of thermal comfort or daylight performance assessments are automatically carried out. The outputs are transformed into consistent, measurable performance indicators aligned with the defined objective functions. Where required, constraint-violation measures are calculated to filter out infeasible solutions prior to optimisation in Phase 03.

Phase 03—Multi-Objective Optimisation

Phase 03 enables systematic exploration of trade-offs among competing performance objectives through a structured optimisation process. In this phase, population-based optimisation algorithms, including evolutionary-based algorithms and swarm intelligence-based algorithms, are employed to iteratively explore the defined design space and evaluate alternative design configurations. In this framework, no explicit weighting is assigned to the objective functions. EC and CC are treated as independent and competing objectives within a Pareto-based multi-objective optimisation approach. This allows the generation of non-dominated solution sets without introducing subjective bias associated with predefined weights. Instead, trade-offs between objectives are explored comprehensively through the Pareto front, enabling decision-makers to select preferred solutions based on project-specific priorities in the subsequent evaluation phase.

Phase 04—Evaluation, Decision-Making, and Knowledge Extraction

This phase, comprising three stages, advances beyond conventional optimisation workflows by extending analysis beyond Pareto-front generation. Post-processing techniques, including extreme-point analysis, balanced-solution identification, clustering and MCDA, are applied to support decision-making under practical constraints. In Stage 11, optimisation outcomes are validated through one or more approaches, such as case studies, code compliance checks, and comparisons with the existing literature, thereby enhancing the credibility of the results. Finally, the translation of optimisation results into design rules of thumb bridges the gap between computational optimisation and practical application, addressing the need for practitioner-oriented outputs in BDO research.

4.3. ICCO-BD Framework Validation

As presented in

Table 9. Results of framework validation by interviewees.

Validation Category	Respondent								Mean	Median	SD	IQR
	1	2	3	4	5	6	7	8
Overall structure of the framework is well-organised	5	4	5	5	4	5	5	5	4.75	0.46	5	0.25
Framework content is logical and coherent	5	5	4	5	5	5	5	5	4.88	0.35	5	0
Process flow is clear and easy to follow	4	4	5	5	4	4	5	5	4.5	0.53	4.5	1
Framework captures all essential stages of BDO	4	4	5	4	5	4	5	5	4.5	0.53	4.5	1
Integration of carbon and cost considerations is appropriate	5	4	4	4	5	5	5	5	4.62	0.52	5	1
Framework is applicable to real-world building design practice	5	4	4	5	5	4	5	5	4.62	0.52	5	1
Framework is suitable for early-stage design decision-making	5	4	4	5	4	5	5	4	4.5	0.53	4.5	1
Framework can support low-carbon and cost-effective design outcomes	5	4	4	5	4	4	5	5	4.5	0.53	4.5	1

, the validation results confirm that the proposed ICCO-BD framework provides a systematic and comprehensive approach to building design optimisation through the explicit integration of upfront embodied carbon (EC) and construction cost (CC) considerations. The mean scores ranged from 4.50 to 4.88, indicating a high level of expert acceptance across all evaluation criteria. The highest mean score was achieved for the logical coherence of the framework content (4.88), followed by the overall structure of the framework (4.75), demonstrating that experts perceived the framework as well-organised, comprehensive, and methodologically robust. Similarly, the integration of carbon and cost considerations and the applicability of the framework to real-world building design practice achieved highly favourable evaluations (mean = 4.62), highlighting the practical relevance and decision-support capability of the framework. The process flow, framework completeness, suitability for early-stage design decision-making, and support for low-carbon and cost-effective design outcomes achieved mean scores of 4.50.

The standard deviation values ranged from 0.35 to 0.53, indicating a relatively low level of dispersion in expert responses and, consequently, a strong degree of agreement among participants. The lowest standard deviation was observed for the logical coherence of the framework content (SD = 0.35), suggesting near-unanimous agreement regarding the clarity and consistency of the framework structure. Slightly higher standard deviations were recorded for the process flow, framework completeness, suitability for early-stage decision-making, and support for low-carbon and cost-effective design outcomes (SD = 0.53). Although these criteria exhibited marginally greater variation in expert perceptions, the overall dispersion remained low, demonstrating strong consensus regarding the framework’s relevance and applicability.

The distribution analysis further confirmed a high level of consensus among experts. The framework content exhibited the strongest agreement (IQR = 0.00), indicating an exceptionally high level of consensus among experts regarding its logical coherence. The remaining evaluation criteria recorded interquartile range (IQR) values of 1.00 or below, suggesting limited variability in expert opinions despite differences in professional backgrounds and practical experience. The observed variability is therefore interpreted as reflecting contextual differences in industry practice rather than disagreement regarding the framework’s overall value. Collectively, the mean scores, standard deviations, and distribution statistics demonstrate strong expert consensus and support the reliability and validity of the proposed ICCO-BD framework.

Despite the highly positive evaluations, the experts also identified several areas for further refinement and practical implementation considerations. With respect to the framework structure, Expert 2 suggested incorporating Indoor Air Quality (IAQ) as an additional performance evaluation criterion within the simulation and optimisation process. This enhancement would enable the framework to consider occupant health and indoor environmental quality alongside embodied carbon and construction cost objectives, thereby providing a more comprehensive assessment of building performance.

Furthermore, Experts 5 and 8 recommended explicitly defining design constraints and performance boundaries based on relevant building codes and regulatory requirements. In response to this suggestion, the framework was refined to emphasise the incorporation of code-compliance requirements within the optimisation process, ensuring that the generated design solutions remain both practically feasible and compliant with applicable regulations.

Regarding practical implementation, Experts 3 and 5 highlighted that, although the framework is highly suitable for early-stage design decision-making, its effective application requires specialised knowledge in optimisation techniques, programming, and building design. Consequently, the implementation of the framework may present challenges for architects, engineers, and other construction professionals who possess limited expertise in optimisation methodologies and computational modelling. The experts noted that successful application of the framework may require the involvement of optimisation specialists within the design team, particularly for large and complex projects where the benefits of optimisation can be fully realised. However, both experts emphasised that this limitation is not inherent to the proposed framework itself but rather reflects the current capability gap within the construction industry regarding optimisation knowledge and skills.

4.4. Limitations of the Proposed Conceptual Framework

Despite its contributions, the ICCO-BD framework is subject to several limitations. Its effectiveness depends heavily on the availability, accuracy, and consistency of input data, particularly for embodied carbon and construction cost parameters, which may vary across regions and project contexts. The reliance on parametric modelling and multi-objective optimisation introduces computational complexity and may require specialised expertise, potentially limiting its applicability in resource-constrained or time-sensitive projects. Furthermore, the optimisation outcomes may be sensitive to the choice of algorithms, parameter settings, and objective formulations, which can influence the resulting Pareto solutions and associated trade-off interpretations. Although the integrated end-to-end workflow improves methodological coherence, its practical implementation may require considerable initial setup effort. Finally, as the framework is conceptual, it has not yet been empirically validated across diverse building typologies and contexts, highlighting the need for future real-world implementation and benchmarking.

5. Conclusions, Limitations and Future Directions

This study developed the ICCO-BD (Integrated Upfront Carbon and Construction Cost Optimisation for Building Design) conceptual framework through a systematic review of 41 peer-reviewed studies and validation by eight academic and industry experts. The review revealed that although BDO research is methodologically mature, there is a strong methodological concentration on OC and global cost, while upfront EC and CCs remain comparatively underexplored in carbon-oriented BDO. These gaps informed the development of a structured four-phase framework that integrates problem formulation, parametric modelling, multi-objective optimisation, and post-processing into a coherent end-to-end workflow. The ICCO-BD conceptual framework contributes theoretically by formalising optimisation as a sequential and interdependent process rather than an isolated algorithmic exercise. It explicitly aligns objective functions, design variables, constraints, and algorithm selection criteria, thereby enhancing methodological transparency and robustness. By embedding baseline benchmarking, structured algorithm selection guidelines, and systematic Pareto evaluation techniques, the framework strengthens the reliability and interpretability of optimisation outcomes.

To further ensure the framework’s validity and practical relevance, it was evaluated by eight experts, comprising industry professionals in building design and architecture and academic researchers in building optimisation, against nine predefined criteria. The evaluation results indicate that the framework is logical, coherent, and well-structured, with strong potential for practical application in supporting early-stage decision-making for low-carbon and cost-effective building design optimisation. The ICCO-BD framework contributes by structuring optimisation as a coherent end-to-end workflow that integrates problem formulation, parametric modelling, multi-objective optimisation, and evaluation. This integration provides measurable benefits by reducing iterative redesign loops, thereby improving time efficiency, enhancing solution quality through more comprehensive Pareto front exploration, and minimising inconsistencies between stages. Furthermore, by embedding systematic trade-off evaluation within the workflow, the framework improves the clarity and robustness of decision-making. Consequently, the proposed framework not only advances the conceptual understanding of BDO processes but also offers practical improvements in optimisation performance and reliability. Moreover, the developed conceptual framework contributes to the literature by systematically integrating upfront EC and CCs within a structured four-phase optimisation process, including objective-driven variable mapping and algorithm selection criteria. While the framework provides a systematic foundation to support early-stage decision-making, it remains conceptual in nature and has been developed primarily based on secondary data sources. Its initial validation was limited to expert evaluations focusing on clarity, completeness, and perceived feasibility, rather than empirical performance testing. Future research should focus on implementing and validating the proposed framework through real-world building projects and optimisation case studies across diverse building types and project contexts. Furthermore, the quantitative superiority of the framework should be evaluated by comparing its performance outcomes with those of existing frameworks reported in prior studies. This can be achieved through benchmarking against established models, enabling a systematic assessment of improvements in key performance indicators such as cost efficiency, carbon reduction, and optimisation effectiveness.

Moreover, the ICCO-BD framework enables early-stage designers to balance upfront EC and CCs while maintaining operational feasibility. However, as the framework has been developed based on secondary data and validated primarily in terms of its clarity, completeness, and feasibility through expert interviews, further empirical validation is required, as it remains conceptual in nature. Therefore, future research should focus on testing the applicability of the framework through real-world case studies, including different building projects, integrating localised carbon and cost datasets, and engaging industry stakeholders to refine and prioritise algorithm selection criteria. In addition to the above, although the proposed ICCO-BD framework was validated by eight experts representing both academic and industry perspectives, the findings should be interpreted within the context of this relatively small expert panel. Therefore, future research should further validate and refine the framework through engagement with a larger and more diverse group of experts from different disciplines, industry sectors, and geographical regions to enhance its generalisability, robustness, and practical applicability. Furthermore, although data saturation was considered during the qualitative validation process and no substantially new insights emerged after the sixth interview, a formal saturation test was not conducted. Future studies could incorporate a formal saturation assessment to further strengthen the rigour and credibility of the framework validation process. While the ICCO-BD framework provides a structured approach for simultaneously considering EC and CCs during BDO, it does not prescribe a specific mathematical normalisation or trade-off mechanism between these objectives. Future research should therefore focus on implementing the framework within real-world optimisation models to investigate appropriate objective normalisation techniques and decision-making strategies for balancing carbon and cost trade-offs. In addition to the above, the regional variability of construction cost data was not explicitly incorporated within the scope of this study, as the framework is developed conceptually. However, cost parameters are inherently location-dependent, influenced by factors such as labour rates, material prices, and market conditions. In practical implementation, the framework can be adapted to different geographic contexts by integrating region-specific cost databases. While the absence of regional calibration may affect absolute cost values, it does not compromise the underlying optimisation structure. Future work should therefore focus on incorporating geographically specific datasets to assess the robustness and transferability of the framework across different regions.

Another limitation of this study is the reliance on a single database. While the use of Scopus alone may introduce coverage and selection bias, its comprehensive scope, combined with the large number of retrieved records, ensures a robust and reasonably representative dataset for the purposes of this review. Nevertheless, the findings should be interpreted with consideration of these limitations, and future reviews may benefit from incorporating multiple databases to enhance coverage and reduce potential bias.