Investigating the Performance of Kolmogorov–Arnold Network on Acoustic Leak Detection in Water Distribution Networks

Water loss due to leaks in water distribution networks (WDNs) represents a significant challenge requiring advanced solutions. Recent studies have increasingly focused on developing machine learning (ML) based leak detection systems. However, such methods are frequently constrained by inherent limitations, including overfitting and suboptimal parameter efficiency. This study examines the potential of Kolmogorov-Arnold networks (KANs) for acoustic leak detection, comparing its performance to convolutional neural network-1D (CNN-1D), CNN-2D, and artificial neural network (ANN). The findings reveal that KAN-related models achieve competitive validation and out-of-sample accuracy, outperforming CNN-2D and ANN while slightly underperforming compared to CNN-1D. Furthermore, KAN-related models demonstrate parameter efficiency when integrated with the deep learning (DL) framework, although they require longer training durations. Experiment results demonstrate the potential of KAN as a reliable tool for leak detection while highlighting the importance of further improving its practical effectiveness. This study advances ML-based leak detection studies and proposes novel hybrid architectures to enhance the effectiveness of leak detection systems.