MACHINE LEARNING APPLICATIONS IN CONSTRUCTION SAFETY LITERATURE

Abstract

Despite significant improvements in safety management practices, the construction industry remains among the most unsafe industries. Thus, it is an essential need to reduce the number of construction accidents through prediction models. In this context, machine learning (ML) methods are extensively used in construction safety literature to predict several outcomes of construction accidents. This study provides a literature review in ML applications in construction safety literature to illustrate research directions for future research. Based on the literature review, 43 journal articles were deeply investigated, and distribution of the articles were classified based on six features: journal, year, adopted machine learning methods, model development approach, utilized dataset, and sub-topics. The findings show that the prediction models in construction safety have taken considerable attention recently. Besides, linear regression and logistic regression were used as a benchmark model, while support vector machine and decision tree were the most frequently implemented ML methods. The number of publications that considered classification problem is two times higher than those adopted regression models. Utilized data were mainly captured from national databases or construction companies. Severity evaluation of construction accidents was the most widely investigated sub-topic, while there is a gap in the literature related to effects of culture on accident outcome and conflict, claim and nonconformance. The findings of this study can provide valuable information for researchers with trends in construction safety literature.