Uncovering Critical Causes of Highway Work Zone Accidents Using Unsupervised Machine Learning and Social Network Analysis

Abstract

Highway work zones are essential for the preservation and improvement of the national road system. Nevertheless, these areas are reported to be among the most hazardous workplaces. Thus, it is crucial to develop appropriate measures to effectively mitigate the safety risks, which require a good understanding of the critical causes of accidents. While there are many previous studies on critical causes of construction accidents, none of them was specifically focused on highway work zones. This type of construction workplace has its own characteristics (e.g., near-passing traffic), which can lead to a unique set of critical causes of accidents. This study used text mining to extract root causes from a large narrative data set of construction accidents at work zones obtained from the Occupational Safety and Health Administration (OSHA). The study applied latent Dirichlet allocation (LDA) modeling on the text corpus to extract 12 root causes, which were subsequently classified into five groups: management, human, unsafe behavior, environmental, and material factors. In addition, social network analysis (SNA) was conducted to gain further insights into the interrelations between the root causes to determine their criticality degree. As a result, four highly ranked causes were identified: supervision dereliction of duty, weak safety awareness, poor construction environment, and risk-taking behavior. The findings of this study offer a new understanding of critical factors that highway agencies and contractors should focus on when developing construction accident prevention strategies at work zones.