Performance improvement for pipe breakage prediction modeling using regression method

Abstract

In water distribution systems, the challenge was always to develop reliable models for predicting the failures for each individual pipe in their lifetime to keep the system reliable. The ability of predicting the pipe failure is one of the most important fundamentals for the effective rehabilitation strategy. Statistical methods are the most used techniques in this field. The prediction accuracy reflects the reliability of the proactive rehabilitation strategies. This article introduces a simple technique to improve and enhance the accuracy of non-linear multiple regression prediction models. The method proposed was applied to predict the number of pipe breaks using 7 predictors for actual water distribution system, where a performance improvement of 4.6% was yielded to the prediction model. Key words: Pipe breakage, multiple regression analysis, residuals, water distribution systems.