Predicting Wall Thickness Loss in Water Pipes Using Machine Learning Techniques

Abstract Wall thickness loss in water pipes has been found to be positively correlated with pipe failure. The reliability of reduces as their wall increases. Although previous studies have investigated failure modeling using historical data, however, indirect via is yet explored. Hence, this study develops machine learning (ML) models predict pipes. Random Forest (RF) and Gradient Boosting Machine (GBM) are chosen the base integrated Bayesian Optimization (BO) algorithm for hyperparameters selection. predictive evaluated root mean square error (RMSE), absolute (MEA), percentage (MAPE), coefficient determination (R 2 ). Based on evaluation metrics, hybrid (i.e., RF+ BO GBM+BO) outperformed (RF GBM), showing importance systematic selection hyperparameters. best model + BO) achieved an RMSE, MAE, MAPE, R value 3.212, 2.494, 11.506, 0.910, respectively. These metrics show high capability model, which can used by infrastructure management