Employing Machine Learning with Optimized SVM Parameter to Detect Water Pipe Leakage

Most products, such as drinking water, oil, gas, and other petrochemical products, are distributed through pipe lines which are considered economical and safe. Regardless the fact, pipe line distribution is susceptible to leakage that could bring about loss to business institutions and pollution to surrounding environment. This study is focusing on the detection of water pipe leak using flow rate parameter. Testing is carried out on dataset format, both using direct data and already processed data format using volume balance method in detecting leakage. The teaching of machine learning needs to highlight some critical phases such as pre-processing, normalization method, and the applied kernel function method that enable to improve the prediction accuracy performance of both size and location of pipe leak. The data collected shows that size accuracy of leak of SVM model for raw dataset is 96.07% at min-max normalization, and for processed dataset is 69.71% at min-max normalization. For location classification, on the other hand, leak accuracy for raw dataset is 84.5% without normalization, and for processed dataset is 75.24% at z-score normalization. The use of optimized kernel function at z-score normalization has shown accuracy improvement in both size and location classification. The accuracy of leak size for raw dataset has improved to 98.25% and for processed dataset to 89.7%. For location classification, the improvement shows 88.96% for raw dataset and 74.42% for processed dataset respectively.