Improving project cost forecasts: an analysis of combining unsupervised and supervised machine learning algorithms with real project data

Purpose – This study aims to develop a new hybrid algorithm that combines supervised and unsupervised ML techniques to improve the precision of predictions for project progress and cost performance measures. Design/methodology/approach – This study utilizes a hybrid project cost forecasting methodology by combining supervised and unsupervised machine learning algorithms. Findings – Computational results of this study demonstrate the superiority of the XGBoost and Random Forest algorithms, both ensemble methods, and confirm the accuracy of this forecasting methodology. Originality/value – Unlike earlier studies that rely on artificially generated data, this study uses a dataset of 117 real-world projects to test the new forecasting technique. The clustering approach to the cost dataset, a novel contribution of the current study, demonstrates enhanced prediction accuracy. Cluster-aware estimate-at-completion forecasting is studied as a process-level decision-support tool.