Poster: An Extreme Value Theory Approach to Anomaly Detection (EVT-AD)

We introduce a new approach to anomaly detection based on extreme value theory statistics. Our method improves detection accuracy by replacing binary feature thresholds with anomaly scores and by modeling the tail region of the distribution where anomalies occur. It requires no optimization or tuning and provides insights into results. This work describes the Extreme Value Theory-Anomaly Detection (EVT-AD) algorithm and provides simulation results for two challenging problems: insider threat and credit card fraud. In these experiments, EVT-AD substantially outperformed a standard threshold-based anomaly detection algorithm, providing accurate detection with few or no false alarms even for scenarios with weak indicators. The results suggest that EVT-AD may offer an improvement over existing statistical methods for security-related problems. Keywords-anomaly detection; extreme value theory; insider threat; credit card fraud