Intelligent sensor data analysis through hybrid deep hierarchical clustering for anomaly detection

Anomaly detection (AD) ensures the integrity and security of industrial control systems (ICSs), such as water distribution systems (WDSs). This research introduces a hybrid approach that combines hierarchical clustering and deep learning (DL) techniques to enhance the effectiveness of AD within ICS. Conventional AD methods often encounter difficulties when dealing with intricate, high-dimensional data and identifying subtle anomalies. By capitalizing on the strengths of hierarchical clustering and DL, our proposed approach aims to surmount these limitations and enhance the performance of AD. To assess the viability of the proposed method, we employed the Water Distribution testbed (WADI) and the BATtle of the Attack Detection ALgorithms (BATADAL) datasets. These datasets encompass a diverse array of normal and attack events, rendering them appropriate for testing the resilience of our hybrid system. As for the BATADAL dataset, the suggested approach reaches an average recall, precision, and F1-score of 0.91, 0.915, and 0.915 consecutively. Correspondingly, the WADI dataset records an average recall, precision, and F1-score of 0.70, 0.81, and 0.67, respectively.