Unsupervised Anomaly Detection in Water Distribution Networks Using One-Class SVM

The deployment of real-time sensors has significantly enhanced water quality monitoring in distribution networks, allowing for continuous data collection. These monitoring tools generate large amounts of highly imbalanced data, dominated by normal operating conditions, with rare but critical anomalies. Detecting these anomalies is essential for identifying potential contamination events, infrastructure failures, or sensor malfunctions. This study explores the use of unsupervised learning, specifically the one-class support vector machine (OC-SVM) for anomaly detection of water quality data in distribution networks. It emphasizes that, with proper data pre-processing and parameter tuning, OC-SVM can provide an effective solution for identifying rare deviations in water quality. Tested on real-world data, the model achieved an F1-score of 0.81 and a misclassification error below 4%, highlighting its effectiveness.