A hybrid fuzzy-logic clustering framework for transparent credit risk assessment in microfinance lending

Abstract

Relevance. The assessment of credit risk for short-term unsecured loans in microfinance institutions is a critical problem for financial stability and portfolio management. This segment is characterised by unstable client behaviour and limited credit history, creating challenges for traditional statistical and machine learning scoring models. Transparency and explainability are increasingly important due to regulatory requirements and the need for adaptive methods capable of handling incomplete and noisy data. This study aims to develop a methodology for applying fuzzy logic to enhance the accuracy, stability, and transparency of automatic credit decisions in the microfinance sector. The main objectives include constructing an interpretable fuzzy model, defining fuzzy sets for key variables, developing a transparent rule base built on clustering and BadRate profiles, and validating internal and external consistency, monotonicity, and transparency. Methods. The study uses a dataset of 601,092 issued and closed loans of a Ukrainian microfinance institution in 2024. The proposed approach combines clustering-based segmentation with fuzzy rule-based credit risk assessment. Clustering was performed using the K-Means method, with the optimal number of segments determined by the Elbow method. Fuzzy sets were constructed using trapezoidal membership functions, and rules were formed based on cluster combinations and normalised BadRate coefficients. Aggregation and defuzzification were performed using the Mamdani method to obtain a numerical RiskScore. Validation included checks for monotonicity, rule consistency, and decision transparency at the observation level. Results. The model identifies stable behavioural segments reflecting different risk levels. The RiskScore demonstrates strong consistency with empirical BadRates distributions and forms three interpretable risk categories: Low, Medium, and High. The proposed fuzzy model effectively predicts credit risk under incomplete and unstable data. Conclusions. The scientific novelty of the study lies in the integration of clustering, BadRate profiles, and rule-based inference into a unified interpretable scoring framework for credit risk assessment in microfinance lending. The proposed approach has practical significance, providing a transparent and interpretable decision-support tool for credit risk evaluation, applicable to risk management, portfolio monitoring, loan loss provisioning, and regulatory-compliant scoring systems.