Μηχανική μάθηση για την αξιολόγηση και βαθμολόγηση του πιστωτικού κινδύνου
Machine learning for credit risk assessment and scoring
View/
Keywords
Credit scoring ; ΤΕΙΡΕΣΙΑΣ Α.Ε. ; AI ACT ; ΓΚΠΔ ; Machine learning ; Black-boxAbstract
The credit risk assessment process is a critical process in the financial sector, enabling banking institutions to assess the probability of default and make informed lending decisions. Traditionally, credit ratings are based on statistical models which, although effective, have notable limitations in terms of accuracy and flexibility, especially as data sources and borrower behaviour evolve. In Greece, the most widely used credit rating system is that of the interbank company known as "TEIRESIAS S.A.". The main purpose of the "TEIRESIAS" information system is to consolidate the financial system and protect commercial credit.
However, this thesis explores the application of machine learning (ML) in credit rating assessment, as ML introduces a transformative approach to credit risk assessment, allowing for greater prediction accuracy and the ability to integrate large, complex, and diverse data sets. It delves into data challenges, including feature engineering, preprocessing, and handling unstructured data.
Emphasis is placed on the new provisions introduced by the EU Regulation on Artificial Intelligence and on model evaluation metrics adapted for credit scoring. Ethical issues, such as bias mitigation, are examined to ensure that ML-based credit models are aligned with legal standards. Finally, the study examines the future of machine learning in credit scoring, including innovations in data provision and real-time risk assessment.
Overall, this diplomatic thesis aims to provide a comprehensive overview of how ML can improve credit risk assessment, creating a more equitable and data-rich financial environment.