Predicting cost elements of construction projects using supervised machine learning techniques

Abstract

Accurately predicting construction project costs remains challenging due to their dynamic and complex nature. While traditional methods address most cost components based on resources it consumes or how such activities perform, certain elements like fuel often rely on expert judgment for validation or adjustment, as traditional methods frequently fail to capture all influencing project parameters. This research explores the feasibility of utilizing supervised Machine Learning (ML) techniques to predict these volatile cost elements, focusing specifically on fuel, a key project cost. The study addresses key gaps identified in the literature, particularly the need for models that can manage the uncertainty of specific cost elements and incorporate a broader range of influencing factors, including macroeconomic parameters. By leveraging historical data extracted from Enterprise Resource Planning (ERP) systems, alongside additional project attributes such as average fuel price and construction cost indices, this study demonstrates a novel, data driven approach to cost estimation. The methodology involved data preprocessing to ensure quality and consistency, followed by feature selection to identify the most relevant attributes influencing fuel cost. Several supervised ML models were compared, to identify model with superior performance. The chosen model was further optimized through iterative refinement techniques, to enhance its predictive accuracy and stability. The findings highlight the potential of supervised ML to revolutionize construction cost estimation practices, offering a more data driven, accurate, and efficient method for managing project budgets realistically.