A hybrid machine learning approach for customer valuation and decision-making in peer-to-peer lending

Risk-Adjusted Revenue (RAR) is a key customer valuation metric that incorporates risk factors into financial assessments. While traditional methods have focused on structured data to estimate RAR, the potential impact of unstructured textual data has not been fully explored. This study addresses this gap by extracting information from 126,000 peer-to-peer (P2P) online loan descriptions using Topic Modelling techniques. The extracted textual data is integrated into both individual and hybrid Machine Learning (ML) models to predict and explain customers’ RAR. A key contribution of this research is the application of Artificial Intelligence by designing and implementing hybrid ML models in finance, combining both structured (hard) and unstructured (soft) features. Unlike conventional credit risk assessment models, which often overlook the value of textual data, this study demonstrates how the incorporation of loan description text can significantly enhance predictive accuracy, customer segmentation, and overall valuation. Our results show that hybrid ML models, particularly those using Latent Semantic Analysis (LSA) for topic-based customer clustering, outperform individual algorithms, achieving a predictive power of R²=97.55%. These models enable the creation of customer portfolios with distinct risk and return profiles, offering valuable insights for financial institutions seeking more accurate customer valuation and segmentation. The findings suggest that integrating textual data into RAR prediction models can provide substantial improvements in both prediction performance and decision-making in the P2P lending market.