Explainable AI Pipelines for Behavioral Fraud Modeling in Online Retail Environments

With online retail fraud posing growing threat to consumers, explainable AI (XAI) has become increasingly important for transparent and actionable risk assessment. This paper presents an XAI-integrated pipeline for behavioral fraud modeling, fusing supervised ensembles of logistic regression and random forests with unsupervised isolation forests to detect both known and emerging behavioral anomalies, including irregular cart sequences and geolocation inconsistencies. SHAP-based attributions are incorporated to deliver instance-level explanations that enhance auditability and support compliance requirements (e.g., PCI DSS). Using a heterogeneous dataset of 150,000 transaction records, the proposed system achieves an F1-score of 0.93 and reduces false positives and manual interventions by 82% relative to an industry-standard rule-based baseline. The architecture supports offline batch analysis and scalable serverless deployment. Pilot studies indicate potential operational cost reductions driven by decreased review workloads and improved detection efficiency. The open-source implementation fosters iterative community refinements, advocating XAI's role in fortifying e-commerce resilience against evolving threats like synthetic identities.