MACHINE LEARNING-BASED RISK CLASSIFICATION FOR CONSTRUCTION PROJECTS: A COMPARATIVE PERFORMANCE ANALYSIS OF RANDOM FOREST, XGBOOST, AND NEURAL NETWORKS IN CIVIL ENGINEERING

Abstract

This research develops a predictive framework for classifying construction project risk levels using machine learning. The study implements and compares three algorithms Random Forest, XGBoost, and Artificial Neural Networks on a comprehensive dataset of 24 project features including cost, schedule, safety, and environmental metrics. Results demonstrate that XGBoost achieved superior performance with 95.5% accuracy and 0.995 AUC, significantly outperforming other models in classifying challenging Medium-risk projects. Feature importance analysis identified Safety_Risk_Score and Anomaly_Detected as the most critical predictors. The findings provide construction managers with a robust decision-support tool for proactive risk mitigation and establish XGBoost as the optimal algorithm for multi-class risk prediction in construction projects.