xAAD–Post-Feedback Explainability for Active Anomaly Discovery

Anomaly detection algorithms are widely used across various domains, but they often suffer from high false positive rates and lack of interpretability. This paper introduces xAAD, a novel approach that combines Active Anomaly Discovery (AAD) with the Assist-Based Weighting Scheme (AWS) explainability metric for Isolation Forest-based anomaly detection. Our method enhances model interpretability and reduces false positives by incorporating expert feedback and providing post-feedback feature importance values. We evaluate xAAD on both synthetic and real-world datasets, demonstrating improved performance in terms of binary classification accuracy and root mean square error (RMSE) compared to traditional AWS. The results show that xAAD consistently outperforms the baseline in both simulated and real-world scenarios, suggesting a positive impact on the interpretability of anomaly detection systems across multiple industries.