ARTIFICIAL INTELLIGENCE AND CONSUMER PROTECTION IN DIGITAL FINANCIAL SERVICES

Abstract

This study proposes an Artificial Intelligence-based Consumer Protection Framework (AI-CPF) that integrates adaptive consumer representation learning, Transformer-based sequential behavioral modeling, Graph Neural Network (GNN)-based relational learning, cross-modal feature fusion, Bayesian uncertainty estimation, fairness-aware optimization, and explainable artificial intelligence into a unified end-to-end architecture. The proposed framework is designed to simultaneously model heterogeneous consumer information, temporal transaction patterns, behavioral relationships, and predictive uncertainty while generating transparent and equitable risk assessments. A comprehensive experimental evaluation was conducted using multiple benchmark machine learning and deep learning models under identical training conditions. The evaluation considered predictive accuracy, precision, recall, F1-score, AUC, Matthews Correlation Coefficient (MCC), computational efficiency, uncertainty estimation, fairness, and explanation stability. The experimental analysis demonstrates that the proposed AI-CPF framework consistently achieves superior predictive capability while maintaining reliable uncertainty quantification, enhanced interpretability, and improved fairness compared with conventional approaches. Furthermore, the ablation analysis confirms that each architectural component contributes positively to the overall performance, highlighting the complementary benefits of integrating sequential learning, graph-based representation, Bayesian inference, and explainable artificial intelligence.