Conceptual estimation of construction duration and cost of public highway projects

Abstract

State Highway Agencies (SHAs) and Departments of Transportation (DOTs) allocate their limited resources to thousands of competing projects in multi-year transportation programs using expert judgement for the expected construction costs and durations. Such estimates overlook influencing parameters known in the planning phase and the importance of building reliable databases to support decision making. Meanwhile, it is possible to generate meaningful predictions in early stages of project development based on historical data gathering and analysis. The present research introduces a newly developed method for conceptual cost and duration estimation for public highway projects utilizing an ensemble of machine learning (ML) models and data collected for projects completed between 2004 and 2015 (roads, bridges, and drainage projects). Unlike previous studies, the proposed method includes project parameters that affect construction durations and costs and were not studied simultaneously before. The parameters considered are facility type, project scope, highway type, length, width, location, level of technical complexity, and new parameters pertinent to payment and procurement methods. The developed method was tested using 29 and 56 randomly selected projects, and the results yielded a Mean Absolute Percentage Error (MAPE) of 7.4% and 4.5% for the duration and cost, respectively, which are lower than the estimation errors of methods reported in recent literature. Additionally, the generalization abilities were assessed by the Mann-Whitney test, and the developed method is found to successfully handle diverse projects. Thus, machine learning models can assist agencies in the review process of competing projects from a high-level management perspective to ultimately develop better management execution programs.