DTD-VAE: Disentangled Temporal Dependencies VAE for Credit Risk Prediction

RESUMO

Evaluating customer creditworthiness is crucial for retail banking operations, as it impacts marketing strategies, customer relationship management, and credit risk control. Traditional methods often struggle to capture complex temporal dependencies and extract pertinent information from customer data, essential for accurate risk assessment. Specifically, they fail to differentiate between temporal patterns indicative of credit risk those reflecting general customer behavior or preferences, leading to suboptimal risk predictions. In this study, we introduce the Disentangled Temporal Dependencies Variational Autoencoder (DTD-VAE), an advancement over conventional VAE, designed to disentangle temporal dependencies and distinguish risk-related features from past customer preferences. The feature inference module of DTD-VAE incorporates an autoregressive temporal dependency learning mechanism that adeptly captures temporal dependencies among latent variables, enriching the model's comprehension of inherent data structure. Furthermore, the generative module utilizes an element-wise gating mechanism that assigns independent weights to each dimension of expert models, enabling a finer-grained disentanglement particularly relevant to credit risk prediction. Extensive experiments on six real-world datasets demonstrate the proposed framework consistently outperforms existing methods, achieving performance gains of 3.2%-4.86% in ROC-AUC and 6.41%-9.71% in Accuracy Ratio.