Clash Relevance Prediction Based on Machine Learning

Abstract

Building information modeling (BIM) has been widely used for clash detection, which has greatly improved the coordination efficiency among multiple disciplines in construction projects. However, the accuracy of BIM-enabled clash detection has been questioned because its outcome includes many irrelevant clashes that have no substantial influence on a project or that can be solved in the subsequent design or construction phases. To improve the quality of clash detection, this paper uses supervised machine learning algorithms to automatically distinguish relevant and irrelevant clashes. This paper selects six kinds of algorithms: J48-based decision tree, random forest, Jrip-based rule methods, binary logistic regression, naïve Bayes, and Bayesian network. The Kruskal-Wallis test was used to compare their performance, and the results found that the Jrip method outperforms the other methods. Finally, a method is provided to identify irrelevant clashes and demonstrate how the clash management process can be improved through learning from historical data.