Data-Driven Injury Severity Prediction by Integrating Clustering Analysis and Deep Neural Network Model

With the rapid development of intelligent transportation systems, injury severity prediction has been paid more attention. This study intended to predict the injury severity with massive data provided. To achieve this goal, the data set was firstly collected from Traffic Accident Database maintained by Nevada Department of Transportation from 2015 to 2017, and then Getis-Ord Gi* was selected as the hot spot analysis. Based on the hot spot analysis, deep neural network was considered to predict the injury severity. By training the data set and comparing the results with deep neural network, the results showed beneficial performance from the goodness-of-fit. Findings revealed that the proposed model can be considered as an alternative to predict injury severity. The results may provide potential insights for reducing the injury severity.