Yahya Nur Ifriza, Tiara Lailatul Nikmah, Subhan Subhan
Peer-to-peer loan represents a financial technology advancement that allows microloans to be distributed online without traditional intermediaries. This system directly connects borrowers and lenders, offering convenience but also exposing lenders to default risks. Loan defaults in P2P lending can lead to substantial financial losses and compromise the system's overall reliability. Therefore, effective risk prediction and management strategies are essential. This study focuses on forecasting loan defaults using a Stacking Ensemble approach enhanced through Random Forest Feature Selection (RFFS). The research utilizes the Lending Club dataset, which consists of 1,961,527 records across 18 attributes. Following feature selection, the top three most influential features were retained for modeling. Performance assessment using a Confusion Matrix indicates that the Stacking Ensemble combined with RFFS achieves superior predictive performance, reaching an accuracy of 99.934%. The findings contribute to the advancement of more refined machine learning techniques for default risk analysis in Peer-to-peer loan platforms. © 2025 IEEE.
Universitas Negeri Semarang, Computer Science Department, Semarang, Indonesia
