DECAF: deep capsule attention network for intelligent auto loan default risk detection

With the increasing challenges of default risk in the auto loan market, traditional risk assessment methods show significant limitations in coping with the rapidly changing market environment. This study proposes an innovative deep learning architecture—Deep Capsule Attention Network (DECAF)—which integrates capsule networks with self-attention mechanisms, effectively enhancing the intelligent detection capability for auto loan default risks. Experimental evaluations of the model’s performance in both conventional and high-risk scenarios demonstrate that the DECAF model achieved an area under the curve (AUC) of 0.924 in the conventional scenario and maintained a high performance of 0.850 in high-risk scenarios. Additionally, DECAF exhibits a significant advantage in terms of reducing false positives (FPR < 0.2), effectively minimizing misjudgments. More importantly, the study reveals that the DECAF model retains high stability even under severe market fluctuations, with only a 7.4% performance drop. These results provide new insights and solutions for financial institutions to optimize risk control strategies in dynamic market environments.