Prediction of Pipes Break in Water Distribution System Using Data Mining Tools “Case Study Nablus Municipality” رسالة ماجستير

Abstract

The problem of water loss from water distribution networks is a major economic problem that worries many stakeholders in the Middle East area, especially those working in municipalities responsible for water distribution networks. Since there is a shortage of water in most countries of the world, this problem occupies great focus in the world especially in major cities, where the socio-economic cost of water loss is increasing. The breaking of pipes in water distribution networks is one of the main reasons for the loss of water from the network, so there is an urgent need to control this problem to prevent water leakage from the pipes by continuous repairing and maintaining the pipes before the break. Therefore, there is a need to analyze and understand the data related to water distribution networks and to use this data in predicting the breaking of the pipelines and identifying the factors and variables that lead to break before broken pipes. using of classical mathematical and statistical tools in identifying the parameters which play a major role in the prediction of pipes’ break patterns is a complex task; because of the complexity of this system so that this research seeks to create an alternative model that is to be used for predicting pipes’ breaks in water distribution networks and for identifying the variables that cause such breaks. In this research, the applied dataset collected from the water distribution system in the Municipality of Nablus, which is one of the larg cities in the northern West Bank area of Palestine that was taken as a case study. The R language was used to implement seven classification models for pipes break prediction depending on three data mining techniques that are Decision Tree, Logistic Regression and Support Vector Machine. II The first three models were built by using one of these three techniques, then four new models have also been built by combining the two of these techniques. Comparing the performance of these models shows that the new model that is built by combining the Logistic Regression and Support Vector Machine techniques, which is called LRSVM model that is most reliable model in the anticipation of pipes' breaks because it gave the best values for most of the calculated performance measures as its error rate varied between 0.01 and 0.12, and it may be able to save up to 0.97 water from the amount of water lost from the network, with an accuracy rate that may reach 0.99.