Σχεδίαση και ανάπτυξη υβριδικού συστήματος αξιολόγησης πιστωτικού κινδύνου με μηχανική μάθηση, επιχειρησιακή λογική και δυναμική βαθμονόμηση πιθανοτήτων

Αναστασοπούλου, Ζωή

Design and development of a hybrid credit risk assessment system using machine learning, business logic, and dynamic probability calibration

Master Thesis

Author

Αναστασοπούλου, Ζωή

Date

2026

Abstract

This thesis addresses the development of an integrated hybrid system for predicting the probability of customer loan default in the banking sector, utilizing machine learning techniques, business rule mechanisms, and data visualization tools. Effective credit risk management represents a critical challenge for financial institutions, underscoring the need for advanced data analysis approaches and intelligent decision support systems. Within the scope of this study, data were collected, explored, and preprocessed to ensure their quality, consistency, and reliability. Subsequently, machine learning techniques were employed for the development of predictive models, which were evaluated using appropriate performance metrics. This process enabled the identification of significant patterns and key factors associated with customer default behavior. Beyond the baseline modeling process, a hybrid prediction mechanism was developed, combining machine learning estimates with business logic rules (rule-based overrides) and dynamic probability calibration techniques. This approach aims to provide a more realistic assessment of credit risk and to translate model predictions into actionable lending decisions. In addition, an interactive data visualization system (dashboard) was developed, enabling users to input borrower characteristics and receive real-time estimates of default probability, risk classification, and recommended lending decisions. Through the use of charts and key performance indicators, the system facilitates the interpretation of findings and supports data-driven decision making. The results demonstrate the substantial contribution of machine learning techniques to the prediction of credit risk, while also highlighting the critical role of business logic integration in improving the practical reliability of predictions. The proposed hybrid approach can serve as an effective tool for supporting customer evaluation processes within the banking industry.

Postgraduate Studies Programme

Πληροφοριακά Συστήματα και Υπηρεσίες

Department

Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών Συστημάτων

Number of pages

103

Language

Greek

URI

https://dione.lib.unipi.gr/xmlui/handle/unipi/19558

Collections

Τμήμα Ψηφιακών Συστημάτων

Show full item record

Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα