How Can a State-of-the-Art Prediction Technique for Random Parameter Models Improve the Construction Work Environment in Ontarian Highways?

Work zones noticeably increase the likelihood of conflicts, collisions, and delays. Sudden changes in the capacity of segments, and aggressive manoeuvres to avoid congestion due to lane closures are the main factors that impact the traffic performance of the work zones. These sudden manoeuvres at the merging point of work zones create an exceptionally hazardous environment for the motorists and workers. The Centers for Disease Control and Prevention (CDC) reported that the average of lethal collisions in the work zone reached 772 in 2015–2017. This value is approximately 200 more than the average of 2008–2014 fatality average. These facts raise the question that how agencies and contractors could predict the hazardous scenarios more accurately and provide a safer environment? One of the most well-established methods to tackle this issue is analyzing the historical data and extracting the most significant factors. Fatality Analysis Reporting System (FARS) data from New York, Pennsylvania, Illinois, and Michigan is used as the basis of the advanced statistical models. This database includes four main categories: Person level information, Pre-crash level information, Vehicle level information, Crash level information. The advantage of using random parameter models is their ability to let the estimated coefficients change across the observations. However, for prediction purposes, it has been always a challenge to select the most appropriate coefficient. This study recommends an innovative technique to overcome this issue by introducing the “similarity level” concept. The prediction results using this method was promising and improved the prediction accuracy by at least 20% in each injury severity category.