Leveraging Large Language Models and Deep Learning for Construction Work Zone Intrusion Risk Assessment

Construction work zones are essential for roadway maintenance. In most cases, the traffic flow should not be interrupted, which requires work zones to be set up alongside the ongoing traffic, endangering the safety of the construction workers. Despite the efforts made to improve safety, statistics indicate increased injury and fatality rates associated with construction work zone workers. While previous studies investigated factors contributing to work zone intrusion by literature reviews and qualitative analyses, such methods heavily rely on subjective interpretation and lack scalability. Therefore, this study aimed to apply natural language processing (NLP) to intrusion reports written by police officers to develop intrusion likelihood and severity evaluation models based on structured tabular and text data. Results showed that aggressive driving and lane changes are the most significant contributors to work zone intrusion. The findings of this study help prevent work zone intrusions and minimize the implications of the intrusions.