Cost Estimating Models for Utility Rehabilitation Projects: Neural Networks versus Regression

Abstract

Due to the poor condition of the sewer and water networks in many communities across the United States, many rehabilitation projects are being undertaken to improve their condition. With limited availability of funds, early and accurate prediction of project costs is highly desirable. Accurate prediction of cost does not only assure allocation of adequate budgets for successful completion but also assists in proper utilization of available limited resources. This paper describes the development of cost estimating models for sewer and water network repair projects. To develop these models, data from a set of 54 projects were used. Data pertaining to these projects were first processed to identify the factors that highly impact the overall cost. These factors were then further processed using two approaches, namely, artificial neural networks and regression analysis, to develop the cost estimating models. A comparison of the accuracy of the predictions from two approaches indicated that the artificial neural network approach provided better accuracy.