Engineering, Geology, Climate, and Socioeconomic Aspects' Implications on Machine Learning-Dependent Water Pipe Collapse Prediction

Abstract

From the impact of several corporeal, mechanized, ecological, and civic conditions, underground water pipelines degrade. A motivated administrative approach of the water supply network (WSN) depends on accurate pipe failure prediction that is difficult for the traditional physics-dependent model to provide. The research used data-directed machine learning approaches to forecast water pipe breakdowns using the extensive water supply network's historical maintenance data history. To include multiple contributing aspects to subterranean pipe degradation, a multi-source data-aggregation system was originally developed. The framework specified the requirements for integrating several data sources, such as the classical pipe leakage dataset, the soil category dataset, the geographic dataset, the population count dataset, and the climatic dataset. Five machine learning (ML) techniques are created for predicting pipe failure depending on the data: LightGBM, ANN, logistic regression, K-NN, and SVM algorithm. The best performance was discovered to be achieved with LightGBM. Analysis was done on the relative weight of the primary contributing variables to the breakdowns of the water pipes. It's interesting to note that pipe failure probabilities are shown to be influenced by a community's socioeconomic variables. This research suggests that trustworthy decision-making in WSN management may be supported by data-directed analysis, which incorporates ML methods and the suggested data aggregation architecture.