Intelligent Pricing Mechanism of Consumer Finance Behavior Based on Multimodal Data Integration

In response to the limitations of traditional consumer-credit pricing models that rely on single-modal demographic and financial tables, this study develops and empirically validates a multimodal intelligent pricing framework that jointly exploits textual declarations, profile images and granular in-app behavioral logs. Each modality is encoded into a 128-dimensional embedding and fused through gated multimodal units and cross-modal attention to capture latent signals of repayment risk. The fused representation feeds a two-stage pipeline in which a deep classifier estimates individualized default probability and a reinforcement-learning policy converts this risk score into compliant, profit-maximising price adjustments. Using 120,000 real loan applications from an international fintech platform, the proposed model reduces mispricing loss by 17 % and lifts AUC to 0.879 compared with the best unimodal and gradient-boosted baselines, while meeting interest-rate caps in five regulatory regimes and sustaining 94 ms online latency. Attention heat-maps and SHAP analyses confirm that pricing decisions remain interpretable and free of single-feature dominance, satisfying emerging AI-governance requirements. The results demonstrate that rich multimodal cues substantially enhance pricing accuracy, fairness and operational robustness, offering a scalable blueprint for next-generation responsible consumer-finance systems.