Comparing Machine Learning Models For Predicting Water Pipelines Condition

Abstract

The majority of water pipelines suffer severe deterioration and degradation challenges. Therefore, this research aims at developing machine learning models that forecast the structural condition of water pipelines. The models are implemented using several techniques, including multiple linear regression, feed-forward neural network, general regression neural network, and support vector regression models. The performance of the aforementioned models is evaluated by measuring the coefficient of determination and root mean squared error using cross-validation. The results show that the general regression neural network model outperforms the other models with respect to the applied metrics. The models are developed using data collected from a water distribution network in Shaker Al-Bahery, Qalyubia Governorate, Egypt. The developed model is expected to assist the water municipality in allocating budget efficiently as well as scheduling of the needed intervention strategies.