A Comparative Study of Explainable AI Techniques in E-Commerce Fraud Detection

E-commerce platforms handle trillions of dollars in transactions every year, yet increasingly sophisticated fraud schemes exploit human behaviour patterns and technological vulnerabilities have also increased in number. While in the context of e-commerce fraud detection, machine learning models have achieved strong predictive performance, the systematic evaluation of the interpretability and performance trade-offs of various XAI algorithms is still limited. This study addresses this gap by conducting a comparative analysis of XAI methods tailored to e-commerce fraud detection scenarios. The study employs a comparative experimental methodology to evaluate three XAI methods in different e-commerce fraud detection scenarios. Using a stratified dataset of transaction records, three XAI techniques—Attention-Ensemble, SHAP-enhanced Random Forest, and LIME-based models—were evaluated across multiple fraud categories Performance was assessed using predictive metrics (accuracy, precision, recall, F1-score, AUC-ROC) and explanation quality metrics (interpretability, complexity, usefulness, actionability). The results from the analysis shows that Attention-Ensemble has the highest Precision (0.941), the highest Accuracy (0.993), the highest Recall (0.897), the highest F1-Score (0.905), and the highest AUC-ROC score (0.978). Similarly, in each of the evaluation metrics, SHAP-enhanced random forest models outperformed the LIME-based methods. Hence, the benefits of LIME’s comprehensibility can be applied to fraud analyst training and client communication, while the increased consistency of SHAP explanations makes risk assessment processes more reliable.These findings demonstrate that hybrid use of XAI techniques can balance predictive accuracy with interpretability, strengthening fraud detection workflows and enhancing trust in AI-driven e-commerce systems. The study contributes to the advancement of transparent, accountable, and actionable fraud detection frameworks in digital commerce.