Artificial Intelligence-Driven Risk Assessment in Civil Engineering Projects: A Comparative Analysis of Machine Learning Models

Successful completion of civil engineering projects involves complex interactions among various components such as managing costs, maintaining schedules, ensuring structural stability, and minimizing environmental impacts. Accurate risk prediction at the project level is important to support efficient resource planning and prevent potential delays or disruptions. Since traditional risk assessment methods based on expert judgments involve lengthy and subjective procedures, the development of data-driven and automated estimation systems is at the forefront. In this study, a comparative analysis of various machine learning methods such as RF, SVM, XGBoost, CatBoost, and LGBM for project risk assessments in civil engineering is presented. The dataset used in the study includes information on project type, cost change records and planning, and real-time history, as well as environmental data and safety documentation, and structural health measurements. Precision, accuracy, sensitivity, and F-score metrics were used to evaluate the models along with ROC-AUC measurements. Experimental studies showed that CatBoost and LGBM outperform the compared models with 95.5% accuracy.