Predicting Traffic Accidents Through Heterogeneous Urban Data: A Case Study

With the urbanization process around the globe, traffic accidents have undergone a rapid growth in recent decades, causing significant life and property losses. Predicting traffic accidents is a crucial problem to improving transportation and public safety as well as safe routing. However, the problem is also challenging due to the imbalanced classes, spatial heterogeneity, and the non-linear relationship between dependent and independent variables. Most previous research on traffic accident prediction conducted by domain researchers simply applied classical prediction models on limited data without addressing the above challenges properly, thus leading to unsatisfactory performance. This paper, through a case study, presents our explorations on effective techniques to address the above challenges for better prediction results. Specifically, we formulate the problem as a binary classification problem. For each road segment in each hour, we predict whether an accident will occur. Big data including all the motor vehicle crashes from 2006 to 2013 in the state of Iowa, detailed road network, and various weather attributes at 1-hour granularity have been collected and map-matched. We evaluate four classification models, i.e., Support Vector Machine (SVM), Decision Tree, Random Forest, and Deep Neural Network (DNN). To tackle the imbalanced class problem, we perform an informative negative sampling approach. To tackle the spatial heterogeneity challenge, we incorporate SpatialGraph features through Eigen-analysis of the road network. Results show that employing informative sampling and incorporating the SpatialGraph features could effectively improve the performance of all the models. Random Forest and DNN generally perform better than other models.