Integrating Explainable Artificial Intelligence in Anomaly Detection for Threat Management in E-Commerce Platforms

In modern cybersecurity, detecting sophisticated threats in high-volume and dynamic network environments is a persistent challenge. Current advanced techniques, such as recurrent neural networks and transformers, achieve remarkable precision but lack interpretability, limiting their adoption in operational environments where analysts must validate predictions and make informed decisions. This gap between precision and explainability creates significant barriers in critical environments such as e-commerce platforms, were real-time traffic and evolving threats demand speed and transparency. This study proposes an explainable anomaly detection model that integrates SHAP, LIME, and Grad-CAM to address these challenges and provide interpretable predictions. The model was evaluated in a representative e-commerce environment, targeting sophisticated threats such as distributed denial of service attacks and unauthorized access. The proposed system enables analysts to understand, validate, and refine model predictions efficiently by combining advanced detection capabilities with real-time explanation generation. The results demonstrate the model’s effectiveness, achieving 95% precision and generating explanations in less than 500 milliseconds, outperforming traditional methods such as clustering and achieving precision comparable to transformers with significantly lower inference times. In addition, the system improved the efficiency of analysts’ decision-making, reducing validation times by 33% and increasing confidence by 20%.