Predictive Risk Management in Construction Projects Using Machine Learning: A Decision Tree and Logistic Regression Approach

Abstract

This thesis investigates how machine learning can improve risk management in construction projects, especially for predicting cost overruns and schedule delays. It uses questionnaire data collected from 243 construction professionals and applies two interpretable models: Logistic Regression and Decision Tree. The study finds that both models can classify projects into low-risk and high-risk categories with very strong internal performance. A key finding is that approved change orders are a major indicator of severe project risk, with projects having eight or more change orders classified as high risk. The thesis also shows that questionnaire-based data can be useful when formal project records are incomplete or unavailable. Overall, the study provides a practical and transparent approach for using machine learning as a monitoring and decision-support tool in construction risk management.