Text Mining-Based Approach for Identifying Critical Accident Causes in Highway Construction

The construction sector is among the most hazardous workplace where workers are more likely to be at risk than in other jobsites. Generally, construction accident reports possess a wealth of empirical knowledge in the form of text summarizing related events. However, analyzing this information source typically requires the high cost of manual content analysis as accident reports are often voluminous and presented in an unstructured format. Text mining and machine learning have recently been applied to extract and leverage valuable information from accident reports. However, few studies have focused on automatically identifying accident causes from highway construction incident reports. This study aims to extract highway construction critical accident causes from a large narrative dataset obtained from the Occupational Safety and Health Administration by adopting the Latent Dirichlet Allocation algorithm. As a result of this implementation, 12 critical accident causes were identified, which were subsequently classified into five groups of accident causes: management factors, human factors, unsafe behavior, environmental factors, and material factors. This study is expected to empower organizations to rapidly analyze and obtain reliable critical highway accident causes from their accident report databases with minimum expert involvement. Managers can use such outcomes to formulate appropriate safety strategies to reduce catastrophes.