An integrated approach of machine algorithms with multi-objective optimization in performance analysis of event detection

Challenges in the provision of a safe water distribution system have become one of the major concerns to the society. Various models and algorithms have been developed so far to incorporate in the early warning systems. This study focuses on the use of machine learning (ML) algorithms on different contaminated datasets. Fine tree (FT) and linear support vector machine (LSVM) were chosen to classify the events. To select the best combination of event and nonevent data, nondominated sorting genetic algorithm-II is integrated with the algorithms that helps to obtain an optimal solution of minimized false positive rate (FPR) and minimized false negative rate (FNR). Results suggest that both FT and LSVM minimized FPR and FNR very effectively. However, FT performed better than LSVM in a supervised and laboratory-controlled dataset, and it showed its superiority in securing robustness over LSVM and fuzziness-based methods in different uncertain scenarios of the study datasets. Moreover, the study initiated a novel approach by executing FT and LSVM models to classify contamination events in a combination of two datasets of various contaminants. It produced better results compared to the Pearson correlation-Euclidean distance (PE) method applied in the same dataset. In addition, the ML algorithms showed their consistency in detecting most of the simulated events using different ranges of spikes.