Research on Dynamic Credit Profit Forecasting Model Based on Customer Behavioural Uncertainty

Conventional risk models rely heavily on static indicators, such as external credit scores (EXT_SOURCE), and thus fail to capture complex dynamics hidden in customer behaviour. To address this limitation, this study introduces a novel uncertainty indicatorentropyto enhance traditional credit risk assessment models in forecasting customer profit potential. This paper applies a Dirichlet Process Mixture Model (DPMM) to cluster customer behaviour and to calculate entropy values based on posterior probabilities, thereby quantifying behavioural uncertainty. We then employ a Generalized Additive Model (GAM) to evaluate the nonlinear effect of entropy on profit rates, controlling for conventional variables such as loan amount, income, and external credit scores. Empirical analysis confirms that entropy has a statistically significant and nonlinear impact on profit rates, challenging the linear assumptions of traditional models. More importantly, our dynamic decision framework reveals that entropy-based models can identify customer groups with both high uncertainty and high profit potential, which are frequently overlooked by conventional credit risk models. These findings provide theoretical support for reframing risk management into a broader strategy of profit maximization and customer value optimization. Despite the studys reliance on a simplified proxy profit function, the results are robust and provide a valuable tool for financial institutions seeking to refine their riskreturn trade-offs.