Work zone crash prediction model and characteristics analysis

This research’s main goal is to improve the safety of work zones. It is commonly thought that work zones have a negative impact on the number of vehicle crashes. In addition to the high cost for maintaining roadway and new roadway construction, safety is one of the highest concerns for local DOTs. This research uses Iowa Crash, INRIX and work zone plan data to predict and analyze key work zone crash characteristics. The work zone crash prediction model was constructed using a negative binomial regression model in combination with a random forest importance plot and an exhaustive search engine. The resulting final data included 511 crashes throughout 32 work zones from 2017-2018. The resulting final model provides a relationship that accurately predicts the number of work zone crashes in Iowa. The equation can predict accurately for data within the boundaries of the variables used in the study in Iowa work zones. The equation had poor accuracy when predicting low risk work zone crashes, or work zones with low crash values. In addition to prediction, the research analyzed the affect each variable in the final model had on the number of crashes. Work zone length was extremely impactful on work zone crashes. While DOT district, divided roadways and AADT were also impactful. Additional research is recommended as work zone data was limited as well as a large proportion was missing. In the near future a similar study is recommended when work zone dates and lengths are more accurately reported. In addition, more work zone data should be used either from other state DOTs or more years of data.