Predicting construction delay risks in Saudi Arabian projects: A comparative analysis of CatBoost, XGBoost, and LGBM

Accurately forecasting construction delay risks is essential for effective project management in the construction industry. This study evaluates the performance of three machine learning algorithms CatBoost, XGBoost, and LGBM in predicting construction delays for Saudi Arabian projects with varying levels of Time Overrun (TO). Model efficiency is assessed using key performance metrics, including accuracy, misclassification error, precision, sensitivity, specificity, false positive rate (FPR), and false negative rate (FNR). The results demonstrate that LGBM outperforms CatBoost and XGBoost across all TO categories. For projects with less than 30% TO, LGBM achieves the highest accuracy and sensitivity. In the 30%-60% TO category, it continues to excel, particularly in accuracy and sensitivity. For projects exceeding 60% TO, LGBM maintains its lead with superior accuracy, sensitivity, and specificity. These findings establish LGBM as a highly effective algorithm for predicting construction delays, offering valuable insights for risk assessment and management in the construction sector. While CatBoost and XGBoost also deliver strong performance, showcasing their ability to predict delays across various TO levels, LGBM's consistent superiority highlights its robustness and reliability. This study advances the application of predictive analytics in construction delay risk management, paving the way for the development of effective risk mitigation strategies.