Adaptive Feature Interaction Model for Credit Risk Prediction in the Digital Finance Landscape

Abstract

The rapid evolution of online financial markets and shifting personal consumption patterns have led to an increasingly complex landscape of credit options. This expansion has heightened the necessity for robust credit risk assessment models, particularly in the face of rising defaults caused by economic downturns and unemployment. This study evaluates various credit risk models, proposing an adaptive feature interaction model to address limitations in conventional frameworks, such as feature interaction oversight and noise interference. Leveraging techniques like a SENET-based gating mechanism and an attention-augmented Multilayer Perceptron (MLP), the proposed model captures deeper feature interactions, yielding improved prediction outcomes. Experimental results reveal the model's superiority across key performance metrics, including AUC, accuracy, and geometric mean, highlighting its potential in enhancing credit risk forecasting. The findings demonstrate a promising application of advanced deep learning techniques in financial risk management, offering more accurate and reliable predictions.