Integrating BIM and Machine Learning for Energy and Carbon Performance Prediction in Office Building Design

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,

I wanted to reach out and express my sincerest appreciation for your efforts in writing this paper.

This MS presents a framework for predicting energy and carbon performance of office buildings during early design stages when detailed information is limited. The study combines BIM and ML to develop predictive models that address the challenge of assessing building sustainability without extensive simulation effort. The authors generated 210 parametric office building models for Orlando, Florida using Design of Experiments methodology, varying 14 design parameters including geometry, envelope properties, HVAC systems, and operational schedules. The framework shows adaptability across different climate conditions and validated successfully against measured data from two university buildings.

The manuscript is well-developed, with clear density and purpose. It follows a classical structure that eases navigation. In my opinion, the proposed framework constitutes a meaningful contribution to the field, enabling architects and engineers to evaluate energy and carbon implications early in the design process, when modifications are most feasible, without requiring detailed simulation for each design iteration.

Major Comments

• If I understood it correctly, the study addresses only 6 of the 19 ASHRAE climate zones, limiting the applicability of results across all US climate conditions. Extending coverage to all 19 zones would strengthen the study. If this was not feasible, the authors should provide rationale for the selected zones.
• The LR (which cover roughly references 2-26) includes 10 citations older than 2019 (42% of total). Given the rapid advancement in this field, a more updated foundation would strengthen the manuscript. I recommend revising this section to incorporate recent developments.
• Table 14 shows that predicted EUI for UCF Global is approximately double the measured EUI, yet the authors state in line 667 that "the RF model reproduced energy trends reasonably well." This characterization appears accurate only for UCF Research. Please provide justification for this discrepancy and clarify the validation outcomes.
• Also in Table 14, please confirm whether EUI units are kBtu/ft²·yr and ensure all tables display units consistently.
• The Methodology is detailed and provides sufficient information for reproduction. The experimental design is sound, though the visual documentation could be enhanced with Revit ecosystem screenshots.
• The Discussion acknowledges relevant limitations effectively. The Conclusions align with the evidence and arguments presented. However, the Conclusion would benefit from a brief explanation of future research directions.
• I found no issues with references or excessive self-citation and I have no ethical concerns with this study.

Minor and Technical Comments

• The MS lacks sufficient visual material, particularly images showing the building model configuration, spaces, and elements.
• Figure 1 requires improved quality. Consider using current colored ASHRAE climate zone maps for better clarity.
• Figures 2-6 needs to be larger for readability. If these are screenshots, apply AI resampling (e.g., Upscayl software) to enhance image quality. Additionally, provide a brief interpretation guidance for the color coding (blue versus orange) to help interpret the results.
• I suggest to include charts to illustrate simulation outcomes, as tables with multiple numerical values, while beautifully done, are difficult to interpret.

Recommendation

This is solid work that merits publication following revisions addressing the points outlined above. Really well done.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The specific comments:

[1] The titles of Sections 1.5 and 2.1 are duplicated; please revise them.

[2] In section 2.1, the model incorporates 11 variables, while only 210 samples are used for model training. What is the basis for this? Does it meet the sample size requirements of each algorithm? This point should be elaborated on in the paper.

[3] The paper fails to present the model or diagram of the research object; it is recommended that this be supplemented.

[4] In section 2, the paper proposes the establishment of a BIM-ML workflow. It is recommended to add a flow chart for intuitive illustration, instead of relying solely on textual description.

[5] In section 2.1.1, this part discusses the selected design variables and their values, as well as the settings of relevant fixed parameters. What is the basis for the selection or setting of these parameters? It is recommended to provide relevant references or explain how they were determined.

[6] In section 2.1.1, building envelope structures (e.g., external walls, roofs, external windows) are also important parameters affecting energy consumption and carbon emissions. Why are they not considered here? How were they set during the simulation?

[7] In section 2.4, the data include 5 performance outcomes. It is recommended to specify how these performance metrics are calculated or what their constituent elements are.

[8] In section 3.1, the paper adopts 4 algorithms to construct prediction models and compares their predictive performance. However, each algorithm has different parameter settings, and parameter changes may also affect the prediction results. At present, the paper does not reflect the parameter settings of each algorithm and the impact of uncertainty caused by parameter variations.

[9] For Table 8 and Table 9, it is recommended to conduct an importance analysis for all 5 performance indicators. In addition, only the ranking is presented—what is the significance of deriving this ranking? No relevant content addressing this is provided in the subsequent text.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper aims to predict energy and carbon performance at early design stage of office buildings. Specifically, it integrates BIM (with simulated datasets) and machine learning (comparing four methods were compared) for the prediction. The paper is clearly written. Some comments for improvements.

1. A table is needed to clearly compare existing approaches of energy and carbon performance prediction. Highlighting the advantages for integrating BIM and machine learning.
2. Please justify the validity of utilizing simulated datasets. Can the simulated datasets reflect all practical situations?
3. Please justify how the 14 independent variables were concluded and selected.
4. The comparison of four regression method can be more detailed.

5. Why deep learning is not allowed for when practical datasets are unavailable.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The paper has been revised according to the comments.

Reviewer 3 Report

Comments and Suggestions for Authors

No further comments