Mapping the Intellectual Structure of Computational Risk Analytics in Banking and Finance: A Bibliometric and Thematic Evolution Study

Abstract

Modern financial practices introduce complex risks, which in turn force financial institutions to rely increasingly on computational risk analytics (CRA). The purpose of our research is to attempt to systematically explore the evolution and intellectual structure of CRA in banking using a detailed bibliometric analysis of the literature sourced from Web of Science from 2000 to 2025. A comprehensive search in the Web of Science (WoS) Core Collection yielded 1083 peer-reviewed publications, which we analyzed using analytical tools like VOSviewer 1.6.20 and Bibliometrix (Biblioshiny 5.0) so as to examine the dataset and uncover bibliometric characteristics like citation patterns, keyword occurrences, and thematic clustering. Our initial analysis results uncover the presence of key research clusters focusing on bankruptcy prediction, AI integration in financial services, and advanced deep learning applications. Furthermore, our findings note a transition of CRA from an emerging to an expanding domain, especially after 2019, with terms like machine learning (ML), artificial intelligence (AI), and deep learning (DL) being identified as prominent keywords and a recent shift towards blockchain, explainability, and financial stability being present. We believe that this study tries to address the need for an updated mapping of CRA, providing valuable insights for future academic inquiry and practical financial risk management applications.

1. Introduction

In academia, it has been well documented over the last century that financial institutions must manage various risks, including credit, market, operational, and systemic risks. Since the first quarter of the 21st century, mainly due to globalization and the complexity of new financial instruments and transactions, it is believed that the banking sector has faced increasing complexity.

It has also been shown that the traditional risk assessment models are not always adequate to handle modern financial data, which has more complex structures and characteristics. In response to the new challenges of the 21st century, research in new technologies utilizing CRA has emerged. The term includes technologies like AI and ML, which can be further analyzed into clusters like data mining (DM), artificial neural networks (ANNs), computational intelligence (CI), operational research (OR), and big data analytics (BDA).

As has also been shown, adopting CRA in banking and finance offers advantages that lead to better financial stability, processing efficiency, and thus competitiveness [1,2,3]. On the other hand, these tools raise concerns about transparency, ethical use, data privacy, and regulatory compliance [4].

Due to the previous work of many researchers, we can spot multiple research streams in the CRA field, including credit risk modeling, corporate bankruptcy prediction, fraud detection, operational efficiency, and Financial Technology (FinTech) innovations [5,6,7]. Over the last decade, academia has produced systematic literature reviews and bibliometric analyses to synthesize the overall research developments and identify emerging trends [8,9,10].

Among early influential reviews, Ref. [5] systematically classified DM methods for financial fraud detection. Additionally, at the early stage of the last two decades of reviews, Ref. [3] published a comparative survey of AI applications in finance, which highlighted ANNs, expert systems, and hybrid models across many financial domains. In another area of research, that of banking efficiency, Ref. [1] approached the research with a bibliometric review of operational research and AI methods applied to bank performance, justifying the relevance of hybrid AI models for operational risk assessment.

Other prominent topics that this preliminary study of the relevant reviews to date has revealed are corporate failure and bankruptcy prediction. Ref. [11] did a comparison of multiple ML models in bankruptcy prediction and found that ensemble methods often outperform traditional statistical models. Ref. [9] emphasized challenges related to interpretability and generalization by systematically reviewing corporate default prediction models. Another relatively recent review by [4] highlighted debates on sampling techniques, model evaluation, and data quality by synthesizing developments in bankruptcy prediction.

FinTech innovations have introduced additional dimensions to CRA research. Ref. [12] reviewed ML-based digital credit scoring models applied to finance, demonstrating how AI supports financial inclusion. More recently, Ref. [13] investigated BDA and Internet of Things (IoT)-based FinTech solutions, emphasizing how these technologies reshape financial services while raising new regulatory concerns. Ref. [14] examined the convergence of AI, cloud computing, and blockchain technologies, identifying opportunities for efficiency and risks related to decentralization. Ref. [15] conducted a comprehensive review of AI-based credit risk assessment, summarizing supervised and unsupervised learning approaches and emphasizing feature selection, model validation, and ethical deployment challenges.

Based on the relevant research and a first comparison on bibliometric analysis and reviews in the field, as shown in Table 1, we believe that many prior reviews have focused relatively narrowly on specific CRA applications or computational techniques, often using limited datasets. This fragmented approach limits the integration of CRA subfields and fails to fully reflect the accelerated expansion of CRA research in recent years. Moreover, few studies adopt comprehensive bibliometric frameworks to systematically map the intellectual structure of CRA research across multiple subdomains.

Table 1. Comparison of previous studies with the present study.

Author Name (Year)	Time Period	Focus of the Study	Database Used	Number of Documents	Methodology	Contribution	Keywords Used
Ngai et al. (2011) [5]	1997–2010	Financial fraud detection via DM	WoS, Scopus	49 articles	Systematic literature review	Classification framework for fraud detection	DM, fraud detection, classification
Amarnadh & Moparthi (2023) [15]	2000–2023	AI-based credit risk assessment methods	WoS	85 articles	Comprehensive review	Overview of AI techniques in credit risk assessment	Credit risk, AI, data science
Fethi & Pasiouras (2010) [1]	1998–2008	Bank efficiency assessment using AI/OR	WoS	196 articles	Bibliometric review	Survey of AI and OR methods in banking	AI, OR, banking efficiency
Veganzones & Severin (2021) [4]	2000–2020	Corporate failure prediction models	WoS	95 articles	Systematic literature review	Updated synthesis of corporate failure models	Bankruptcy, failure prediction, financial distress
Bahrammirzaee (2010) [3]	1994–2009	AI applications in finance	WoS, Scopus	125 articles	Comparative review	Review of ANNs, expert systems, and hybrid models	AI, expert systems, ANN
Barboza et al. (2017) [11]	1995–2014	Bankruptcy prediction with ML	WoS	130 articles	Meta-review	Comparative performance of ML bankruptcy models	Bankruptcy, ML, prediction
Andronie et al. (2023) [13]	2010–2023	BDA management algorithms in FinTech	WoS, Scopus	75 articles	Systematic review	Applications of BDA and IoT in FinTech	BDA, FinTech, AI, IoT
Alaka et al. (2018) [9]	2000–2017	Corporate default prediction via ML	WoS	95 articles	Systematic review	Evaluation of ML vs. classical corporate default models	Corporate default, ML, bankruptcy
Kumar et al. (2021) [12]	2000–2020	Digital credit scoring for rural finance	WoS	63 articles	Literature review	ML applications in rural microcredit	Credit scoring, rural finance, ML
Lazaroiu et al. (2023) [14]	2010–2023	Blockchain-based FinTech management	WoS	45 articles	Systematic review	AI, cloud computing, and blockchain integration	Blockchain, FinTech, AI

To address these gaps, we present a comprehensive bibliometric analysis of CRA research in banking. Using data from the WoS Core Collection, a structured search query was applied, retrieving 2692 records. Following filtering based on publication years (2000–2025), subject areas, document types, and language criteria, 1083 peer-reviewed articles from 383 academic journals were retained, as listed in Table A1 of Appendix A. From these, 62 review articles were extracted for detailed analysis.

The following research questions guided our review:

• RQ1: What are the publication trends in CRA research?
• RQ2: Where are the most influential publications (outlets and articles) on CRA?
• RQ3: Who are the most prolific contributors to CRA research (authors, countries, and institutions)?
• RQ4: What insights do existing research themes and topics provide about CRA?
• RQ5: Has CRA evolved into a multidisciplinary research field bridging computer science, finance, and decision sciences?
• RQ6: What future research avenues can be explored to enrich our understanding of CRA?

We attempt to take a new approach compared to earlier reviews by combining different areas of CRA, like credit risk, bankruptcy prediction, fraud detection, operational efficiency, and FinTech innovations, into a single bibliometric framework. The analysis employs VOSviewer for co-citation and thematic clustering [16] and Bibliometrix in RStudio 4.5.0 for performance analysis (Ref. [17]). The findings provide an updated, reproducible, and comprehensive mapping of CRA research, offering insights into future academic inquiry and practical financial risk management applications.

In order to optimize the applications and benefits of a dynamic financial landscape, further research is needed to address the growing academic interest. This paper has the following structure: Section 2 describes the bibliometric methodology employed and the data collected. Section 3 presents the results. Section 4 provides a comprehensive discussion, outlines implications for academic research, and concluding remarks.

2. Materials and Methods

2.1. Data Collection

2.1.1. The Source of Data Collection

The WoS database was selected for the data collection due to its extensive coverage of high-quality peer-reviewed journals and its robustness for bibliometric analysis. WoS is renowned for its systematic indexing, multidisciplinary orientation, and comprehensive citation tracking, ensuring the accuracy and reliability of bibliometric analyses [18]. While this study relies on the Web of Science Core Collection for its bibliometric analysis, we acknowledge that Scopus and Google Scholar may index additional journals relevant to CRA, including practitioner-oriented outlets. Future research could incorporate multiple databases to achieve broader coverage.

2.1.2. Keyword Selection Strategy and Refinement Process

The keyword selection process was carried out in two main phases to ensure accuracy. In phase one, we examined CI and its further synonyms or analytical terms, banking and finance, and their synonyms or derivative words in online dictionaries to select the most relevant keywords. In phase two, we employed a WoS search for CI terminologies and AI, ML, ANN, banking and finance, credit risk, systemic risk, etc. After refining the query, the search was conducted based on specific inclusion criteria, detailed in Table 1.

While the generic term ‘risk’ covers a broader array of topics, we refined our search to specific risk categories most relevant to computational applications in banking to improve precision and avoid retrieving unrelated research. Future studies might consider broader or alternative keyword combinations, including terms like ‘idiosyncratic risk,’ to explore additional research niches within CRA

The extensive preliminary dataset required refinement to increase relevance and specificity. First, we narrowed the publication period to articles from 2000 to 2025, justified by the limited number of financial CRA studies before 2000. Second, the search was further filtered to include only relevant journals from economics, finance, business, operations research, computer science, engineering, and multidisciplinary fields. Third, we restricted the results to articles published in English-language journals only. Lastly, we included peer-reviewed research articles and reviews, resulting in a refined dataset of 1083 records.

Table 2 summarizes the step-by-step filtering approach and corresponding outcomes.

Table 2. Final query and refinement steps are used for data collection.

No.	Search Query	Results
#1	Final query: TS = ((“computational intelligence” OR “machine learning” OR “artificial intelligence” OR “data mining” OR “deep learning” OR “neural network”) AND (“bank*” OR “financial institution*” OR “credit risk” OR “risk management” OR “systemic risk”) AND financ* AND bank*).	2692
#2	Limiting the study to studies published from 2000 to 2025.	2655
#3	Included only fields of economics, finance, business, operations, computer science, engineering, and multidisciplinary journals.	1120
#4	Limited online to English journals, research articles, and reviews,	1083

2.2. Study Approach and Tools

The bibliometric approach, introduced initially by [19], systematically analyzes the networks of scholarly literature based on citation patterns. This methodology has gained extensive adoption for its ability to produce accurate, transparent, and reproducible results, providing comprehensive insights into intellectual advancements within research domains [20]. Furthermore, bibliometrics effectively explores various dimensions, including journals, topics, authors, institutions, and geographical research distributions [21]. When combined with content analysis, bibliometric methods allow the summarization of contributions to a research field, providing valuable directions for future research.

VOSviewer, developed by [16], is a software tool for constructing and visualizing bibliometric networks, enabling analyses such as co-authorship (to reveal collaboration patterns), co-citation (to identify intellectual foundations of a field), bibliographic coupling (to group documents sharing similar references), and keyword co-occurrence (to discover thematic structures and emerging research trends). These analyses collectively help uncover the intellectual, social, and conceptual landscapes of CRA research. Moreover, the Bibliometrix software developed by [17] and accessed via its user-friendly Biblioshiny interface in RStudio provided an interactive interface that facilitated performance analysis by generating descriptive statistics on publication trends, authorship patterns, and citation metrics. This provides an overview of the productivity and impact dimensions of CRA research. Collectively, these computational tools provided a robust methodological framework for comprehensive science mapping analysis and thematic exploration of the CRA banking literature.

3. Results

3.1. Dataset General Information

Table 3 presents preliminary data and basic metrics. 1083 articles were analyzed, with an average of 19.9 citations per document. These articles used 2906 author keywords and featured 2952 authors, 103 of whom were single authors. The annual growth rate in the CRA domain was 21.64%, with a collaboration index of 3.13 and international co-authorship at 30.38%.

Table 3. Publication and citation metrics for CRA research.

Statistic	Value
Panel A. Publication Metrics
Total publications (TP)	1083
Article	1021
Review	62
Number of active years (NAY)	26
Productivity per active year (PAY)	43.32
Panel B. Citation Metrics
Total citations (TC)	21,556
Average citations per publication (TC/TP)	19.9
h-index	72
g-index	114
Panel C. Co-authorship Metrics
Number of contributing authors (NCA)	3387
Number of unique authors (NUA)	2952
Authors of single-authored publications (ASA)	103
Single-authored publications (SA)	110
Co-authored publications (CA)	973
Collaboration index (CI)	3.13
International co-authorships (%) (ICA)	30.38%

3.2. Publication Trend: A Temporal Analysis (RQ1)

Between 2000 and 2010, CRA-related studies remained minimal, averaging fewer than five articles per year, as illustrated in Figure 1. From 2011 to 2019, output increased gradually, reaching 58 papers by 2019. A significant surge followed, peaking at 143 articles in 2023. The compound annual growth rate for 2000–2025 was 21.64%, highlighting CRA’s transition from an emerging field to a structured and expanding research domain.

Figure 1. Annual scientific production on CRA in banking (2000–2025).

3.3. Publication Outlets (RQ2)

The journal Applied Soft Computing has the highest number of published articles (25), followed by Computational Economics (24), the European Journal of Operational Research (24), Risks (23), and the Annals of Operations Research (21). Furthermore, Table 4 lists the top journals by total citations, also including Scopus and WoS rankings, providing valuable insights for researchers seeking high-impact sources in CRA research.

Table 4. List of top contributing journals to CRA research by total citations.

Source Journal Titles	TC	NP	TC/NP	h- Index	PY_ Start	Scopus SJR Quartile	WoS JCR Quartile
Decision Support Systems	2009	15	133.93	11	2008	Q1	Q1
European Journal of Operational Research	1925	24	80.21	18	2006	Q1	Q1
Applied Soft Computing	1603	25	64.12	18	2008	Q1
Knowledge-Based Systems	1524	19	80.21	18	2006	Q1	Q1
Neural Computing and Applications	937	13	72.08	10	2010	Q1	Q1
International Journal of Bank Marketing	513	21	24.43	10	2015	Q2
Information Sciences	508	14	36.29	10	2007	Q1	Q1
Journal of Banking and Finance	476	3	158.67	3	2010	Q1	Q1
International Review of Financial Analysis	460	18	25.56	11	2013	Q1	Q1
Technological & Economic Development of Economy	450	7	64.29	5	2012	Q2	Q3
Research in International Business and Finance	369	10	36.90	7	2018	Q2	Q3
Journal of Forecasting	342	21	16.29	11	2000	Q2	Q2
Journal of Business Research	325	6	54.17	6	2018	Q1	Q1
Journal of Research in Interactive Marketing	321	3	107.00	3	2018	Q2	Q3
Neurocomputing	318	8	39.75	7	2010	Q1	Q1

Note: Source journal titles that received a minimum of 300 citations are shown in the list. TC = total citations; NP = number of publications; h = H-Index; TC/NP = citations per publication; PY = publication start year.

The most cited sources in terms of total citations (TC) are Decision Support Systems (2009), the European Journal of Operational Research (1925), Applied Soft Computing (1603), Knowledge-Based Systems (1524), Neural Computing & Applications (937), and the International Journal of Bank Marketing (513). These journals have been instrumental in advancing research at the intersection of computational intelligence and banking risk analytics, consistently publishing influential studies on algorithmic modeling, predictive analytics, and data-driven financial decision-making.

3.4. Globally Cited Articles (RQ2)

Table 5 presents the most globally cited articles on CRA in banking, providing valuable insights into the most impactful research in this field. The most cited article, by [5] in Decision Support Systems, highlights the application of DM techniques for financial fraud detection through a structured classification framework and comprehensive literature review.

Table 5. Most globally cited articles on CRA in banking.

Sr. No.	Article Title	1st Author (Year)	Source Title	Cited by
1	The application of data mining techniques in financial fraud detection: A classification framework and an academic review of literature	Ngai EWT (2011) [5]	Decision Support Systems	595
2	A data-driven approach to predict the success of bank telemarketing	Moro S (2015) [22]	Decision Support Systems	492
3	Assessing bank efficiency and performance with operational research and artificial intelligence techniques: A survey	Fethi MD (2010) [1]	European Journal of Operational Research	452
4	Consumer credit-risk models via machine-learning algorithms	Khandani AE (2010) [2]	Journal of Banking & Finance	357
5	A comparative survey of artificial intelligence applications in finance: artificial neural networks, expert system and hybrid intelligent systems	Bahrammirzaee A (2010) [3]	Neural Computing & Applications	301
6	Prediction of financial distress: An empirical study of listed Chinese companies using data mining	Geng R (2015) [7]	European Journal of Operational Research	294
7	Financial distress and bankruptcy prediction among listed companies using accounting, market and macroeconomic variables	Tinoco MH (2013) [6]	International Review of Financial Analysis	238
8	Predicting financial distress and corporate failure: A review from the state-of-the-art definitions, modeling, sampling, and featuring approaches	Sun J. (2014) [8]	Knowledge-Based Systems	230
9	Does bank FinTech reduce credit risk? Evidence from China	Cheng MY (2020) [23]	Pacific-Basin Finance Journal	224
10	Deep learning models for bankruptcy prediction using textual disclosures	Mai F (2019) [24]	European Journal of Operational Research	212

Another highly cited paper by [22] focuses on predicting the success of bank telemarketing using a data-driven approach. The study presents a robust predictive model built on real-world banking campaign data. Ref. [1] investigated OR and AI techniques to assess bank efficiency and performance. Ref. [2] examined consumer credit risk models utilizing machine-learning algorithms to enhance prediction accuracy. Ref. [3] found that a comparative analysis of ANN, expert systems, and hybrid intelligent systems provides meaningful insights into AI’s role in financial applications. Ref. [7] examined financial distress prediction among Chinese listed companies, highlighting the effectiveness of DM techniques. Ref. [8] offered a comprehensive review of state-of-the-art methods in corporate failure prediction. These influential articles contribute significantly to developing computational frameworks and data-centric methodologies in banking risk analytics.

3.5. Prolific Contributors, Authors (RQ3)

Table 6 lists the ten most productive authors in CRA research, along with their respective affiliations, countries, total publications (TP), total citations (TC), and the starting year of their publications (PY_start). The authors represent a diverse geographical spread, including China, Taiwan, France, Hungary, the UK, Portugal, and Greece.

Table 6. Top 10 contributing authors of CRA research.

Author	Affiliation	Country	TP	TC	PY_ Start
Sun Jie	School of Management, Harbin Institute of Technology	China	9	646	2008
Tsai CF	Department of Information Management, National Central University	Taiwan	8	843	2009
Li Hui	Nankai University	China	8	646	2008
Shi Yong	University of Chinese Academy of Sciences	China	5	228	2010
Du Jardin Philippe	Edhec Business School	France	5	216	2010
Virag Miklos	Corvinus University of Budapest	Hungary	5	121	2012
Liang Deron	National Central University	Taiwan	4	477	2015
Mues Christophe	University of Southampton	UK	4	215	2012
Ribeiro Bernardete	University of Coimbra	Portugal	4	194	2016
Petropoulos Anastasios	Bank of Greece	Greece	4	102	2020

Note: TP = total publications; TC = total citations; PY_start = year of first CRA publication.

Sun Jie, affiliated with Harbin Institute of Technology in China, leads in TP with nine, accumulating 646 TC. Li Hui, also from China, matches Sun Jie’s citations with eight publications. Tsai CF from Taiwan’s National Central University stands out with the highest TC (843) from eight publications, starting in 2009.

Shi Yong (China), Du Jardin Philippe (France), and Virág Miklós (Hungary) each have five publications, with varying citation counts. Liang Deron (Taiwan), Mues Christophe (UK), Ribeiro Bernardete (Portugal), and Petropoulos Anastasios (Greece) have four publications, with their starting years ranging from 2012 to 2020, indicating more recent contributions for some. The data generally suggest a correlation between higher publications and higher citations, though with notable exceptions like Tsai CF’s impact.

3.6. Prolific Contributors, Countries, and Institutions (RQ3)

3.6.1. Countries’ Scientific Production and Most-Cited Countries

Table 7 provides an overview of the top 10 corresponding authors’ countries in the dataset, listing the number of articles authored through single-country (SCP) and multi-country (MCP) collaborations. The top five countries by total article count are China (212), India (109), the United States (77), the United Kingdom (51), and Spain (46). China (53), the United Kingdom (31), the United States (29), Italy (17), and India (15) exhibit the highest levels of international collaboration in CRA research, as indicated by their MCP values. For relative terms, as a percentage of their TP, we observe a different distribution with the UK (60.78%) coming first, followed by Italy (45.95%), the USA (37,66%), France (33.33%), and Australia (28%), with China and India only coming ahead of Iran, which can be interpreted as small level of international co-authored publications in these countries. Nevertheless, these findings highlight the prominent geographic distribution and underscore the collaborative dynamics driving the global advancement of computational risk analytics in banking.

Table 7. Top 10 corresponding authors’ countries by TP, SCP, and MCP.

Sr. No.	Country	TP	SCP	MCP	MCP_Ratio
1	China	212	159	53	25.00%
2	India	109	94	15	13.76%
3	USA	77	48	29	37.66%
4	United Kingdom	51	20	31	60.78%
5	Spain	46	34	12	26.09%
6	Italy	37	20	17	45.95%
7	France	27	18	9	33.33%
8	Australia	25	18	7	28.00%
9	Poland	25	20	5	20.00%
10	Iran	24	21	3	12.50%

Note: SCP = single-country publications; MCP = multiple-country publications; MCP_Ratio = ratio of internationally co-authored publications.

3.6.2. Affiliations’ Scientific Output

In Table 8, we illustrate scientific output in CRA research of the leading ten affiliations to offer insights into global productivity and historical engagement. Islamic Azad University (Iran) exhibits preeminence in both recent (25 articles in 2025) and cumulative (108 total articles) publication volume, having an initial year of production in 2013. Its articles per year rate (8.3) is surpassed by the Ministry of Education and Science (Ukraine) and the University of London, which demonstrate higher annual outputs of 9.8 and 8.8, respectively, despite both entering seven years later (2020). The geographical distribution underscores a diverse international research landscape, encompassing Asian, European, and African institutions.

Table 8. Top 10 affiliations by scientific production in CRA research.

Rank	Affiliation	Country	2025 Articles	Total Articles	per Year Article	PY_ Start
1	Islamic Azad University	Iran	25	108	8.3	2013
2	Egyptian Knowledge Bank	Egypt	21	54	5.4	2016
3	Min. of Education and Science	Ukraine	21	59	9.8	2020
4	University of London	United Kingdom	17	53	8.8	2020
5	Chinese Academy of Sciences	China	15	97	6.1	2010
6	Bucharest Uni. of Econ. Studies	Romania	14	79	6.1	2013
7	National Central University	Taiwan	14	141	8.3	2009
8	National Institute of Technology	India	14	55	7.9	2019
9	European Central Bank	Eurozone	12	46	7.7	2020
10	University of Granada	Spain	12	50	7.1	2019

Note: The European Central Bank operates across Eurozone countries and is headquartered in Frankfurt, Germany.

Notably, the National Central University (Taiwan) distinguishes itself with the highest total article count (141) and the earliest recorded first production year (2009), signaling a sustained, long-term commitment. Including governmental entities, such as the European Central Bank, alongside academic institutions highlight varied organizational contributions to CRA research. Furthermore, the prominent presence of institutions with relatively recent first-production years (e.g., University of London, 2020; University of Granada, 2019) suggests a rapid intensification of their research efforts in CRA, swiftly elevating them to top-tier status.

3.7. Intellectual Structures of CRA Research RQ4

Moving on from descriptive analysis of the CRA bibliometric data, we continue to analyze more intellectual structures of the field. To do that, we transition to bibliometric research with VOSviewer, a more sophisticated tool for further employing strategies like co-authorship analysis, keyword co-occurrence analysis, bibliographic coupling, and co-citation analysis. Each of these analyses was configured using fractional counting, with specific thresholds for inclusion as reported below. The combination of these methods ensures a multidimensional understanding of the knowledge landscape in computational intelligence and risk management research in financial institutions.

3.7.1. Co-Authorship (Author Collaboration Networks)

Author collaboration networks or co-authorship, as frequently used, analyze the collaboration patterns among authors based on joint publications, with the intention to reveal research communities and leading contributors within the fields of CRA. Our co-authorship analysis was conducted using VOSviewer 1.6.20. To have the broadest spectrum of results, we applied no minimum requirements regarding the number of publications and citations. Out of 2952 authors, 60 were connected, resulting in nine distinct collaboration clusters, as shown in Figure 2.

Figure 2. Co-authorship (author collaboration networks) of CRA in banking.

The most collaborative authors in each cluster were determined by using a hierarchical ranking based on the number of collaboration links, total link strength, number of publications, and citation count. Drawing on the initial dataset, together with co-occurrence maps and thematic overlays, each cluster was also assigned a dominant indicative research subfield as illustrated in Table 9.

Table 9. The most influential co-authorship authors in the collaboration networks of CRA in banking.

Cluster/Color	Author	Key Metrics (Links/Docs/Cit.)	Indicative Subfield
1/Red	Hassan, M. Kabir/Rabbani, Mustafa Raza	12/3/45, 5/2/21	Recent Trends and Applications in Financial AI
2/Green	Chi, Guotai/Habib, Tabassum	11/5/35, 8/3/13	Credit Risk Modeling and Financial Inclusion
3/Blue	Chen, Zhensong/Qu, Yi	7/2/12, 7/2/12	DL and Advanced Predictive Analytics
4/Yellow	Wang, Jing/Long, Xingchen	10/4/51, 4/1/4	Financial Analysis and Decision Making with Advanced Models
5/Purple	Abedin, Mohammad Zoynul/Fahmida-emoula	12/4/84, 5/2/30	ML and Data-Driven Insights
6/Turquoise	Shi, Yong/He, Jing	11/4/228, 3/1/9	Computational Finance and Optimization
7/Orange	Shi, Baofeng/Meng, Bin	8/2/79, 4/1/47	Interdisciplinary Applications of AI and Data Science
8/Brown	Zhou, Ying/Bai, Fengshan	10/4/17, 3/1/0	Emerging Trends and Methodological Innovations
9/Pink	Lu, Yang/Liu, Xiaohui	7/2/40, 5/2/129	Quantitative Methods and Algorithmic Trading

3.7.2. Keyword Co-Occurrence (DE—Author Keywords and ID—Keywords Plus)

Keyword co-occurrence examines how frequently author-defined or algorithmically generated (WoS Keywords Plus) keywords appear together in the same documents, thus identifying dominant themes and emerging topics directly or indirectly emphasized by the researchers.

• Keyword Co-occurrence (DE—Author Keywords)

To identify the thematic structure of research in computational risk analytics (CRA) in banking, a co-occurrence analysis of author keywords (DE) was conducted using VOSviewer. The threshold was set at a minimum of five occurrences, resulting in 121 keywords, of which 117 were connected and visualized in the network. These were grouped into 12 clusters based on their co-occurrence link strength, forming a detailed thematic map as illustrated in Figure 3. Keyword co-occurrence (DE—author keywords) network.

Figure 3. Keyword co-occurrence (DE—author keywords) network.

The most frequent keywords included ML (296 occurrences), AI (157), DL (73), bankruptcy prediction (52), credit risk (41), and financial distress (40). We believe that although bankruptcy and financial distress may seem closely related and are sometimes used interchangeably in the literature, they should be maintained as separate keywords in our analysis. This is because financial distress generally refers to the early stages of financial trouble, with potential avenues for recovery, whereas bankruptcy represents a formal legal process with more severe consequences. Furthermore, in banking research, this distinction is significant because it reflects different levels of systemic risk, regulatory interventions, and modeling approaches in CRA studies.

Each cluster represents a distinct thematic domain:

• Cluster 1 (red) focuses on ML and hybrid models in bankruptcy and credit default prediction, including significant words like ensemble, DT, and Support Vector Machine (SVM);
• Cluster 2 (green) concentrates on AI integration in financial services, featuring keywords like FinTech, digital banking, financial inclusion, and blockchain;
• Cluster 3 (blue) represents advanced deep learning applications in textual data, ESG, and credit analytics;
• Cluster 4 (yellow) captures traditional econometric and scoring methods, including Altman Z-score, logit model, and financial ratios.

The keyword network seen in Figure 3 shows a well-structured and maturing research field, with clear thematic boundaries and growing interdisciplinary integration between finance, computing, and decision sciences.

As far as temporal evolution is concerned (Figure 4), the overlay scores by average publication year indicate an evolution of focus toward more recent concepts such as Extreme Gradient Boosting (XGBoost), blockchain, explainability, and financial stability, especially post-2022.

Figure 4. Temporal evolution of keyword co-occurrence (DE—author keywords) network.

This reflects a shift from traditional statistical modeling to interpretable, robust AI frameworks. Average normalized citation scores also highlight impactful keywords like DL (1.74), financial stability (1.68), and ensemble (1.06).

• Keyword Co-occurrence (ID—Keywords Plus)

Using VOSviewer to continue complementing the DE keyword analysis, a co-occurrence network of Keywords Plus (ID) was generated. With a minimum occurrence threshold of 5, the final map includes 151 terms across 12 clusters, as visualized in Figure 5. Keywords Plus, being algorithmically extracted from titles and cited references, offers a broader and more systemic view of thematic connections in the CRA-related literature.

Figure 5. Keyword co-occurrence (ID—Keywords Plus) network.

The most frequently occurring terms were bankruptcy (227 occurrences), models (218), classification (146), credit risk (41), and financial distress (68).

As we can observe in Figure 5, each cluster group is related to terms that reflect conceptual trends:

• Cluster 1 (bankruptcy, models, classification) maps the computational core of CRA methods, such as default prediction, ensemble models, and ANNs;
• Cluster 2 (banking, impact, management) focuses on financial services and adoption frameworks, including FinTech, information systems, and technology acceptance models;
• Cluster 3 contains structural topics like corporate governance, performance, and corporate social responsibility (CSR), indicating interest in firm-level risk and disclosure;
• Other clusters cover the rest of the areas, such as macroeconomic instability (crisis, monetary policy); sustainability and ESG terms; and computational enhancements like optimization, feature selection, and survival analysis.

As far as temporal evolution is concerned (Figure 6), the temporal evolution of keyword co-occurrence (ID—Keywords Plus) network and overlay scores indicate that recent attention has shifted to topics such as blockchain, big data, internet banking, explainability, and sustainability, especially post-2022.

Figure 6. Temporal evolution of keyword co-occurrence (ID—Keywords Plus) network.

Keywords with high normalized citation scores include acceptance model (2.74), information technology (1.71), and fraud (1.77), revealing impactful and, at the same time, emerging themes.

This co-occurrence map, built on Keywords Plus, is an integrator of how the field of CRA in banking connects technical sophistication with sector-specific relevance and policy considerations.

3.7.3. Bibliographic Coupling (Documents)

Bibliographic coupling measures the similarity between documents based on shared references to detect groups of publications that cite similar literature.

The bibliographic coupling analysis reveals how documents are intellectually connected based on shared references. Applying a minimum citation threshold of 40, the analysis included 138 documents grouped into seven clusters, as shown in Figure 7. This approach identifies clusters of articles that, despite differing in methods or focus, build upon similar bodies of literature.

Figure 7. Bibliographic coupling in documents of CRA.

Each cluster represents a tightly coupled thematic stream.

• Cluster 1 (Red): Bankruptcy Prediction and Hybrid Methodologies

This cluster, having 298 total link strength (TLS), is primarily characterized by research focusing on bankruptcy prediction, often employing advanced techniques such as hybrid models and AI. Documents in this cluster explore various approaches to improve the accuracy and robustness of bankruptcy prediction models. For instance, Ref. [8] investigates the application of hybrid intelligent models for corporate financial distress prediction. Ref. [24] delves into DL models specifically designed for bankruptcy prediction, leveraging textual disclosures. Furthermore, Ref. [25] contributes to this theme by examining corporate bankruptcy prediction using an ANN and decision tree (DT) framework. The cluster also features work on intelligent techniques for financial forecasting, as seen in [26], which reviews AI in financial distress prediction.

• Cluster 2 (Green): Advanced ML for Credit Scoring and Financial Distress

With 417 TLS, this is a prominent and highly active cluster, characterized by the application of advanced machine learning techniques, particularly ensemble methods, to credit scoring and financial distress prediction. It features works exploring various algorithms such as ANN, random forests (RF), gradient boosting (GB), and SVM, exemplified by [3], which provides a comprehensive survey of AI applications in financial distress prediction, and [27], which investigates ML methods for credit scoring. A significant sub-theme is the optimization and hybridization of these models for improved predictive accuracy and robustness, often addressing issues like imbalanced datasets and feature selection in complex financial data, as explored by [28] in a review of the credit scoring literature and [29] in their review of financial credit risk assessment.

• Cluster 3 (Blue): AI and FinTech in Finance

The blue cluster comes third with 146 TLS and focuses on the intersection of AI and FinTech, exploring their transformative impact on various aspects of finance. Research within this cluster examines the integration of AI algorithms and machine learning into financial services, particularly in areas like credit risk assessment and digital transformation. For example, Ref. [30] explores how AI transforms relationships in the financial services industry. Ref. [31] proposes a digital servitization value co-creation framework for AI services, outlining a research agenda for digital transformation in financial service ecosystems. Ref. [32] discusses AI and the digital transformation of financial services, while [33] provides a survey of AI techniques in finance and financial markets.

• Cluster 4 (Yellow): DM and ML in Financial Fraud and Risk

This cluster has 104 TLS and highlights the application of DM and ML techniques for detecting and managing various forms of financial risk, including fraud and systemic risk. The works in this cluster emphasize using computational methods to identify patterns and anomalies in financial data. Ref. [5] provides a comprehensive survey of DM applications in financial fraud detection. Ref. [34] focuses on ML methods for systemic risk analysis within financial sectors. Additionally, [35] offer a review of the literature on DM applications in accounting, providing an organizing framework for the field.

• Cluster 5 (Purple): Feature Engineering and Model Enhancement for Bankruptcy Prediction

In terms of TLS (128), purple is the fourth cluster that specifically concentrates on enhancing the performance of bankruptcy prediction models through advanced techniques such as feature engineering and robust model development, often addressing issues like imbalanced datasets. Ref. [36] conducts a comprehensive study on the role of financial ratios and corporate governance indicators in bankruptcy prediction. Ref. [37] introduces a novel two-stage hybrid financial distress prediction model. The application of sophisticated algorithms to improve predictive accuracy is also a key theme, as seen in [38], which explores AI for credit scoring using financial and non-financial information.

• Cluster 6 (Turquoise): Credit Risk Management and Loan Defaults

The 107 TLS turquoise cluster is dedicated to research on credit risk management, with a particular emphasis on understanding and predicting loan defaults within the financial industry. The studies here often analyze factors contributing to credit risk and evaluate methodologies for its mitigation. Ref. [1] provides a comprehensive literature review of credit scoring models using evolutionary algorithms. Ref. [39] investigates the determinants of mortgage defaults. Furthermore, Ref. [40] examines risk and risk management specifically within the credit card industry, while [41] focuses on building a hybrid model for credit risk evaluation using credit card data.

• Cluster 7 (Orange): Predicting Bank Insolvencies and FinTech’s Impact on Stability

Finally, with 64 TLS, this cluster primarily focuses on the prediction of bank insolvencies and the broader implications of FinTech on financial stability. Research in this area utilizes various predictive models to anticipate bank distress and explores how the advent of FinTech influences credit risk and overall financial system resilience. Ref. [42] examines predicting bank insolvencies using machine learning techniques. Ref. [43] contributes to this by anticipating bank distress in the Eurozone using an XGBoost approach. Additionally, Ref. [23] investigates whether bank FinTech reduces credit risk, providing evidence from China, while [44] discusses the relationship between banking and information technology, highlighting the role of AI and FinTech.

The overlay visualization by average publication year illustrated in Figure 8, shows that clusters related to fairness, explainability, FinTech integration, and ESG risk have gained traction after 2020.

Figure 8. Temporal evolution of bibliographic coupling in documents of CRA.

Meanwhile, as illustrated in Figure 9, the citation overlay map of bibliographic coupling in CRA documents emphasizes that foundational machine learning research in credit and default prediction continues to be heavily referenced.

Figure 9. Citation overlay map of bibliographic coupling in documents of CRA.

Interpreting the most cited articles with the help of these structures, we observe how CRA in banking has evolved from technical algorithm development toward broader financial, technological, and social applications.

3.7.4. Co-Citation (Cited References)

Co-citation analyzes how often pairs of references are cited together across the dataset, aiming to reveal foundational works and the intellectual base of the CRA in the banking field.

The co-citation analysis identifies foundational literature in the field of CRA in banking by examining how frequently references are cited together across documents. Applying a threshold of thirty minimum co-citations, the analysis included fifty-nine references grouped into five intellectual clusters. The results are visualized in Figure 10.

Figure 10. Co-citation analysis. Cited references of CRA.

Key findings per cluster:

• Cluster 1 (red): Foundational Financial Distress Prediction Models. In this cluster, prominently featuring references such as [45,46], Refs. [47,48] epresent the foundational theories and models in financial distress prediction, notably including the Z-score and O-score models. These works are central to the field, evidenced by their high citation counts and substantial inter-cluster links, indicating their pervasive influence across various research streams. References like [49,50] also reinforce this cluster’s focus on early quantitative approaches to corporate solvency assessment. Its high density and central positioning in the network images of Figure 10 underscore its role as a core theoretical pillar;
• Cluster 2 (green): Machine Learning and AI in Financial Prediction. Dominated by works from [11,51,52,53], this cluster signifies the integration of advanced ML and AI techniques into financial prediction. The presence of core ML algorithms like RF and GB highlights a methodological shift towards more sophisticated, data-driven approaches for forecasting financial outcomes. This cluster’s growth, particularly with more recent publications, suggests an increasing emphasis on algorithmic performance and predictive accuracy in CRA;
• Cluster 3 (blue): Knowledge-Based Systems and Expert Applications. Key references within this cluster include [54,55,56]. This cluster reflects research leveraging knowledge-based systems and expert systems for decision support in financial contexts. It often involves the application of AI techniques to structured financial data, aiming to provide actionable insights for risk assessment and corporate analysis. The overlap with machine learning aspects of the second cluster suggests a pragmatic application of predictive models within decision support frameworks;
• Cluster 4 (yellow): Contemporary Expert Systems and Financial Applications. This cluster, featuring [8,9,36,57], signifies more recent developments in the application of expert systems to financial problems. While still focused on expert systems, these references often incorporate newer methodologies or address contemporary financial challenges. Their connections to other clusters suggest a continued evolution and diversification of expert system applications, often integrating with statistical or machine learning paradigms;
• Cluster 5 (purple): Hybrid Models and Decision Support Systems. Including works from [25,58,59,60], this cluster likely represents research focused on hybrid modeling approaches and comprehensive decision support systems. These often combine elements from traditional statistical methods, machine learning, and expert systems to create robust predictive and analytical tools. The interdisciplinary nature of this cluster highlights a trend towards multi-methodological integration to tackle the complexities of corporate financial analysis effectively.

Furthermore, as illustrated in Figure 11, the co-citation density map [45] is highlighted as the most central intellectual node, with enduring influence over decades. In contrast, the second cluster features recent high-impact papers on AI, with more dispersed and evolving citation patterns, suggesting intellectual expansion toward interdisciplinary techniques.

Figure 11. Density map of cited co-citation references of CRA.

This analysis affirms the field’s dual structure: it is anchored in classic financial modeling while rapidly absorbing innovations from AI, optimization, and data science.

4. Discussion—RQ5 and RQ6

Our analysis demonstrates that CRA in banking and finance has evolved into a strong multidisciplinary field, effectively bridging computer science, finance, and decision sciences. This evolution is evident across several study dimensions, from keyword co-occurrence patterns to the intellectual clusters identified through co-citation analysis.

The interpretation of the keyword co-occurrence of the author keyword (DE) network in Figure 3 provides significant evidence of interdisciplinarity. We believe that the prominence of ML (296 occurrences), AI (157), and DL (73), together with financial terms like bankruptcy prediction (52) and credit risk (41), illustrates a core integration of computational techniques within financial risk domains in general. As we previously noted in 3.7.2, we deliberately distinguish between financial distress and bankruptcy to capture the different risk levels, qualities, and modeling approaches relevant to CRA research, especially given the regulatory and systemic implications in the banking sector. According to our view, the first cluster is focused on ML and hybrid models for bankruptcy and credit default prediction. There, we can find words such as ensemble, DT, and SVM, which highlight the direct application of advanced computational algorithms to solve complex financial problems. The concentration on AI integration in financial services, with keywords like FinTech, digital banking, and blockchain, as documented in the second cluster, further underscores the fusion of computer science innovations with traditional banking operations, which can also be illustrated by the temporal evolution as Figure 4 is showing further accentuation of this trend, and a shift towards more recent concepts like XGBoost and explainability, which are inherently computational but are increasingly applied to enhance transparency and robustness in financial decision-making.

Furthermore, as we notice through our analysis, the keyword co-occurrence of Keywords Plus (ID) network in Figure 5 corroborates and expands upon these findings. Keywords Plus terms such as bankruptcy (227 occurrences), models (218), and classification (146) are central to finance and computational methods. Terms like bankruptcy, models, and classification, found in the first cluster, directly map the computational core of all CRA methods, including default prediction and neural networks.

The second cluster, which focuses on banking, impact, and management, emphasizes the integration of IT acceptance models within the financial services sector. The emergence of topics like blockchain, big data, and internet banking, shown in Figure 6 as recent areas of focus, further solidifies the interdisciplinary bridge, indicating that financial challenges are increasingly being addressed with sophisticated data science and distributed ledger technologies.

Moving forward to our analysis, we study the findings of the bibliographic coupling (documents) in Figure 7, which further illustrates how distinct thematic streams converge through shared foundational literature, reinforcing the multidisciplinary nature of CRA. What we found is that the first cluster, briefly described as bankruptcy prediction and hybrid methodologies, demonstrates how research in financial forecasting leverages advanced computational techniques like ANNs and DTs [24].

Furthermore, it also highlights the application of DL models to textual disclosures for bankruptcy prediction, which is a clear interdisciplinary field combining NLP with financial analysis [23]. In the subfield Advanced Machine Learning for Credit Scoring and Financial Distress, represented by the second cluster, there are included keywords that showcase the intensive use of ensemble methods, ANNs, RF, and SVM in credit scoring and financial distress prediction [27,28].

Cluster 3, AI and FinTech in Finance, explicitly highlights the blurring lines between computer science and traditional finance, creating new avenues for research in areas such as digital servitization [31] and the redefinition of financial relationships through AI [30]. In other words, it directly addresses the transformative impact of AI on financial services, including digital transformation and credit risk assessment [30,31,32].

Finally, our study also examined the co-citation (cited references) analysis as seen in Figure 10 and provided crucial insights into the intellectual foundations. While the first cluster described as Foundational Financial Distress Prediction Models is rooted in classical financial models like [45,48], we notice that the second cluster entitled ML and AI in Financial Prediction distinctly showcases the integration of advanced ML and AI techniques, such as those found in [11,51,53]. This field not only applies new technologies but also builds upon established financial theories, as this can be interpreted by the persistent influence of seminal financial works alongside the increasing co-citation of machine learning method papers. To further exemplify the interdisciplinary nature, combining elements from traditional statistics, ML, and expert systems to create robust analytical tools and the hybrid modeling approaches, we noticed the works of [25,54,55,59,60] in the third cluster and the fifth cluster, respectively.

In summary, according to our bibliometric analysis, we found that CRA has integrated tools from computer science at a fundamental level, creating a new interdisciplinary domain. The goal of this integration is to allow for and further research for more sophisticated risk assessment, prediction, and management, leading to enhanced financial stability and decision-making capabilities.

Before moving on to our final research question regarding the future directions of research, it is essential to acknowledge that, while the present study aims to provide a comprehensive mapping of CRA research by including all publication sources, future analyses could be conducted that focus exclusively on high-ranking journals. Such an approach would offer additional insights into the quality dimension of CRA research outputs, complementing the broader perspective adopted here. Also, it is essential to point out that while this study used Biblioshiny and VOSviewer, we acknowledge that tools like CiteSpace, which could not be used due to the advanced features licensing limitations for big datasets, offer burst analyses to detect emerging topics. Future research could incorporate such methods for deeper trend insights.

Based on the observed trends and intellectual structures, we are moving on to our final research question regarding the future directions of academic research in the field of CRA in banking and finance, and we can identify several promising future research avenues that can be explored to further enrich our understanding of the field. As we mentioned earlier, our research highlighted that the evolving interplay between computational advancements and the complex dynamics of the financial sector can be seen as a significant driving force towards research progress in the field.

Our first suggestion has to do with the inherent black-box nature of many deep learning and ensemble methods that pose significant challenges for regulatory compliance, ethical considerations, and practical adoption in banking. Our study highlights a shift towards more complex AI models. On the other hand, as demonstrated in Figure 4, with the term “explainability”, future research should focus on developing and implementing Explainable Artificial Intelligence (XAI) techniques tailored to financial risk models.

Our second suggestion has to do with the integration of CRA with Environmental, Social, and Governance (ESG) factors that present a burgeoning research frontier. As we demonstrated in Figure 6 regarding the temporal evolution of Keywords Plus, increased attention to “sustainability” and ESG terms has been noticed in the last few years, especially post-2022. Future research could focus on enabling banks to evaluate and manage their portfolios better and to align them with ESG objectives and explore how AI and machine learning models can effectively analyze vast amounts of unstructured ESG data, like corporate reports, news articles, social media, and how to identify and quantify ESG-related financial risks.

Continuing to our third suggestion, through our research, we noticed that advanced data analytics for unstructured and alternative data sources are crucial. While the analysis indicates a move towards textual data in bankruptcy prediction (Mai et al., 2019 [24]), we believe that there is significant potential in leveraging other unstructured and alternative data forms, and taking that into account, research should investigate novel deep learning architectures like transformers for text, graph NN for network data, and advanced feature engineering techniques to extract meaningful insights from these diverse data sources.

Another area of interest for future study, we believe, is that of Decentralized Finance (DeFi) and the increasing prominence of “blockchain”. Figure 4 and Figure 6 signify the growing academic interest in this area. As DeFi ecosystems expand, new forms of financial risks emerge, including smart contract vulnerabilities, liquidity risks in decentralized exchanges, and governance risks in Decentralized Autonomous Organizations (DAOs). Future research could try to explore how CRA methodologies can be adapted to assess and manage these unique risks transparently and securely by developing on-chain risk analytics, applying machine learning to detect anomalies in blockchain transactions, and creating predictive models for DeFi protocol failures, thus understanding the interplay between traditional banking and the evolving DeFi landscape, which, in our view, will be critical for future financial stability.

Another emerging intersection, relatively critical for future financial stability, is the one between Regulatory Technology (RegTech) and CRA. Almost twenty years after the last global financial crisis of 2007, financial regulations have become increasingly complex, and there is a growing need for technology-driven solutions to ensure compliance and enhance supervisory oversight. Taking that into account, we believe that future research could focus on how CRA techniques, particularly AI and ML, can be applied to automate regulatory reporting, detect compliance breaches, and enhance supervisory analytics by developing models for real-time risk monitoring, stress testing, and anti-money laundering (AML) efforts that are both efficient and auditable.

Finally, the dynamic interaction and collaboration of human intelligence and AI in the context of decision support systems in CRA require further investigation. In our view, although AI models can provide sophisticated predictions, the ultimate and final responsibility for risk management should lie with human decision-makers, as they can combine the different kinds of human intelligence with a human sense of art. Optimizing the interaction between human experts, decision makers, and AI systems in CRA with further exploration into designing effective decision support systems that integrate AI insights with human intuition; studying the impact of AI on human cognitive biases in risk assessment; and developing training programs for financial professionals to utilize CRA tools effectively are also avenues of future research exploration.