Infrastructure projects frequently encounter budgetary and schedule overruns. A cost overrun refers to the variance between anticipated and actual project expenditures, whereas a schedule overrun denotes the divergence between the actual duration of a project and the timeframe originally stipulated in the contract. This paper presents a predictive algorithm utilizing a gradient boosting model to forecast potential cost and schedule overruns during the project's earliest phase. The findings indicate the optimal hyperparameters for the Gradient Boosting Regressor, comprising 0.01 learning rate, a max tree depth of 3, and 300 estimators. The optimized model demonstrates commendable performance in predicting conceptual cost and schedule overruns, with R² approximately 0.96, RMSE approximately 5.68, MSE approximately 32.32, and MAE approximately 3.84, thereby affirming the suitability of data-driven machine learning techniques for forecasting cost and schedule overruns in infrastructure projects. The study demonstrates that early identification of potential overruns enables project managers and policymakers to implement proactive mitigation strategies, such as contingency planning, resource reallocation, and schedule optimization, while practitioners can utilize these strategies for real-time performance evaluation and risk mitigation.

Keywords

Cost overrun, Forecasting, Gradient boasting regressor, Infrastructure projects, Schedule overrun