Comparative Analysis of Machine Learning Algorithms for Clash Relevance and Risk Level Prediction in BIM Design Coordination

As construction projects become increasingly complex, the number of clashes between building systems rises. While existing software systems automatically detect clashes, they often generate excessive irrelevant clashes and lack the capability to prioritize them by risk level. This study evaluates the performance of multiple machine learning (ML) algorithms in clash relevance and risk level prediction using different attribute combinations in building information modeling (BIM). Among the tested algorithms, Random Forest performed the best in predicting clash relevance across all attribute combinations, while Gradient Boosting demonstrated the highest accuracy in risk level predictions. Notably, clash relevance prediction achieved higher accuracy than risk level prediction. The comparative analysis advances the body of knowledge by highlighting the performance of ML techniques in predicting clash relevance and risk level using diverse attribute combinations. Furthermore, this study offers practical recommendations for selecting optimal ML algorithms to support BIM clash management.