Predicting Vulnerable Road User Crashes Based on Seasonable Pattern

Pedestrians, bicyclists, and motorcyclists are called vulnerable road users (VRU) because they lack the protection from hard shell and safety belt. In recent years, they make up over 30% traffic fatalities in the U.S. Since VRUs are exposed to ambient environment, their activities follow outdoor weather conditions and exhibit strong seasonal patterns. Understanding the seasonal patterns of VRU crashes and being able to accurately predict their future occurrences can help decision-makers to allocate safety resource by season and achieve higher return in investment. This study used four data-driven models to analyze the 2007 to 2016 VRU crashes in Pennsylvania, starting from the easy to follow proportion and projection model, then delving into the seasonal autoregressive integrated moving average model, the local level model, and the exponential smoothing state space model. They all capture the annual trends and seasonal patterns of VRU crashes well. This study emphasizes methods of predicting full year VRU crashes from partial year crash data based on knowledge of seasonal patterns. The results show good prediction can be achieved by just knowing the first five months' crash data.