Early warning system for highway construction projects using GA-SVM

Abstract

Construction project changes vis-a-vice their impact on the cost and time performance has been the main concern of practitioners and scholars for decades; hence, many attempts have been made to develop early warning mechanisms for decision-makers. Since many factors drive project changes, various strategies can be adopted to mitigate further alterations after awarding the contract. This study aimed at developing multiple Support Vector Machine (SVM) classifiers to enhance change prediction accuracy. 5,628 completed construction-oriented projects datasets comprising preconstruction information, obtained from the Indiana Department of Transportation (INDOT), were used to develop the models. All free kernel parameters were determined using Genetic Algorithm (GA) optimization. The results show that the models built using SVM classifiers with Radial Basis Function (RBF) kernel function were successful in decreasing the uncertainty associated with change order occurrence in the early phases of the projects. Therefore, the study recommends that agencies, developers, and contractors store historical project data to build early warning systems using machine learning techniques.