Forecasting Completed Cost of Highway Construction Projects Using LASSO Regularized Regression

Abstract

Finishing highway projects within budget is critical for state highway agencies (SHAs) because budget overruns can result in severe damage to their reputation and credibility. Cost overruns in highway projects have plagued public agencies globally. Hence, this research aims to develop a parametric cost estimation model for SHAs to forecast the completed project cost prior to project execution to take necessary measures to prevent cost escalation. Ordinary least square (OLS) regression has been a commonly used parametric estimation method in the literature. However, OLS regression has certain limitations. It, for instance, requires strict statistical assumptions. This paper proposes an alternative approach—least absolute shrinkage and selection operator (LASSO)—that has proved in other fields of research to be significantly better than the OLS method in many respects, including automatic feature selection, the ability to handle highly correlated data, ease of interpretability, and numerical stability of the model predictions. Another contribution to the body of knowledge is that this study simultaneously explores project-related variables with some economic factors that have not been used in previous research, but economic conditions are widely considered to be influential on highway construction costs. The data were separated into two groups: one for training the model and the other for validation purposes. Using the same dataset, both LASSO and OLS were used to build models, and then their performance was evaluated based on the mean absolute error, mean absolute percentage error, and root mean square error. The results showed that the LASSO regression model outperformed the OLS regression model based on the criteria.