Traffic Crash Prediction Based on A Time Series Method

According to Texas Motor Vehicle Crash Statistics, there have been no deathless days on Texas roadways ever since 2003. In 2020, the fatality rate on Texas roadways increased by 18.94% compared to the data in 2019. The total estimated economic loss of all Texas motor vehicle crashes in 2020 is $43,400,000,000. To accomplish the goal of lowering the state’s traffic fatality rate and the total number of traffic fatalities and injuries, local transportation agencies are in urgent need of novel crash prediction models that can improve the accuracy of future crash prediction. To fill this gap, this study applies a time series method – Autoregressive Moving Average (ARMA) models – to predict the number of crashes in Travis County.  In this project, the monthly number of crashes from 2010 to 2019 are obtained from the Texas Crash Records Information System (CRIS) database. The first nine-year crash data (2010 – 2018) are used to develop prediction models. Then, the predictions from the models are compared with observed crash data in 2019 to evaluate the performance of the prediction models. The results indicate that the developed ARMA models can identify the major trends in the motor vehicle traffic crash data and predict the number of crashes with tolerable prediction errors.