Traffic Accident Predictive Model for Better Allocation of Resources in Qatar

Qatar’s economy has proliferated recently, with an unprecedented population expansion from 1.2 million in 2012 to 2.95 million in 2021. The booming economy led to a rapid increase in vehicle counts, especially in urban areas. As a result, traffic accidents have become a significant issue, which causes injuries, damaged properties, and, occasionally, deaths. Even though traffic accident prediction is critical to saving lives and avoiding loss caused by vehicle and material damages, it is a complex task. The main challenges are related to the interference of several factors within the spatial-temporal settings, such as human behavior, weather conditions, availability of real-time traffic data, and the complex traffic environment.

To address these challenges, this research proposes a new intelligent traffic police dispatching system that proactively predicts the occurrence of traffic accidents in each zone in the subsequent police shift and hence takes responsive action to send out patrol vehicles accordingly. To accomplish this target, the traffic accident data from 2017 to October 2023 in 98 zones in Qatar has been obtained. Then, we innovatively proposed the TrafficTransformer for multidimensional, multi-step traffic accident prediction.

The results of this work have shown that several factors play a role in traffic accident prediction in Qatar, which is linked to the nature of the zone and different daily, weekly, and annual timings. Moreover, the proposed model efficiently addressed the central gap in the reported tools in the literature, which failed to simultaneously model dynamic spatiotemporal correlations of the data. Comparative experiments and interpretability analysis were also conducted on a real-world data set. The results indicate that our model can not only yield superior prediction performance but also has the advantage of interpretability.