Optimasi Penilaian Risiko Kredit BNPL melalui Integrasi Algoritma K-Means Clustering dan Random Forest untuk Segmentasi Pengguna di Ekosistem Fintech

Abstract

Buy Now Pay Later (BNPL) services have revolutionized digital credit access, yet they present new challenges in credit risk management due to the limitations of traditional financial data. This study aims to optimize credit risk assessment by integrating machine learning techniques, specifically K-Means Clustering for customer segmentation and Random Forest Classifier for default prediction (default_flag). Data were analyzed using Principal Component Analysis (PCA), which successfully retained 74.70% of the data variance to validate the cluster structure. The study identified three unique segments: The Safe Strategists, The High-Risk Youth, and The Affluent Late-Payers. The implementation of the Random Forest model demonstrated stable performance with an accuracy rate of 70% and a weighted recall of 70%. Feature Importance analysis revealed that digital behavioral variables, such as repayment_delay_days and app_usage_frequency, are significant predictors that compete with conventional indicators like credit_score. These findings emphasize the importance of transforming from static credit scoring to dynamic behavioral scoring to mitigate systemic risks within the Fintech ecosystem.