Application of Gradient Boosting for Predicting Cost Overrun in Construction Projects

Abstract

Cost overrun remains a major challenge in construction project management, particularly in regions characterized by geographical constraints, logistical complexity, and high project uncertainty. Conventional statistical approaches often struggle to capture nonlinear relationships among factors associated with construction cost overruns. This study proposes a hybrid framework integrating Structural Equation Modeling–Partial Least Squares (SEM-PLS) and Gradient Boosting to predict a latent Cost Overrun construct in construction projects in Southwest Papua Province, Indonesia. SEM-PLS was employed to assess the validity and reliability of the research constructs, comprising 24 predictor indicators across three explanatory constructs and three outcome indicators for the latent Cost Overrun construct. The validated indicator-level latent scores were used as input features for the Gradient Boosting model, while the latent Cost Overrun construct was used as the prediction target. The model achieved a mean absolute error (MAE) of 1.7838, a root mean square error (RMSE) of 2.2079, a mean absolute percentage error (MAPE) of 19.10%, and an R2 of 0.6173. Feature importance analysis indicated that Risk Factors contributed most strongly (53.64%), followed by Social Factors (26.03%) and Geographical Factors (20.33%). The proposed framework provides a structured approach for risk-oriented construction cost management in high-risk regions.