Credit Risk Assessment System Based on Deep Learning: A Systematic Literature Review

The credit risk classification of clients is an essential task in financial institutions, as it allows the responsible officer to grant or deny a loan to a potential borrower. Therefore, the resulting score requires precision. However, this process has limitations, including insufficient diversity of methods, complexity of algorithms, available data, among others, which hinder the work of those involved. The objective of this study is to determine the techniques and machine-learning algorithms developed to assess debtor credit risk through a systematic literature review using the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) methodology. The aim of this meta-analysis is to determine the recent studies in the field of financial risk assessment in order to identify the algorithms and recent metrics. Research papers published between 2018 and 2023 were collected and analyzed. The databases consulted were Scopus, Springer, Science Direct, and Taylor and Francis. A comprehensive analysis was conducted, resulting in, the selection on 53 most relevant articles. The results show an evolution from traditional machine learning to deep learning or hybrid algorithms. Moreover, the latter has better precision capabilities compared than other algorithms used for borrower risk assessment. Finally, accuracy (98.60%) and Area under the Receiver Operating Characteristic (ROC) curve (AUC) (97%) are the highest values identified by the authors in the reviewed research using deep learning compared to existing algorithms.