Towards Predicting Traffic Accident Consequence Using Transformer Learning and Clear AI Technologies

In today’s societies, Road traffic accidents are emerging as some of the major threats that put the lives of people as well as the economy at risk. The strategy of addressing such risks for this study is novel as it proposes the use of Transformer Learning coupled with Explainable AI. We use MobileNet architecture to predict the accident severity level with approximately the best possible accuracy of 98 percent. MobileNet, commonly used for efficient and other and embedded vision applications, has been adjusted and trained for the severity of the road accident prediction. The research method is built upon the Transformer model that can be applied to determine dependencies in the dataset on accidents as well as provide sound predictions. The use of Explainable AI enhances the understanding of the readers on the proposed model and reveals the relationship between the variables that influence the level of an accident. Such transparency is important to the readers, government institutions as well as traffic management authorities as they can then make the necessary adjustments. It involves data handling of an extensive record of past accidents and seeks to incorporate several variables that may range from climate during a particular accident to the type of roads, to the time of the day the accident occurred. The results of our training and testing are well tuned and checked so that we achieved the prediction accuracy of 98% regarding the probability of an accident and its severity indexes. Furthermore, the Explainable AI part reveals such characteristics influence prediction outcomes and aligns the model characteristics with real-life circumstances; as well as the high efficiency this research reached in terms of predicting the severity of the road traffic accident, it brings AI techniques that make this process fully automated yet easily explained. Therefore, this research is useful in designing safer roads and in aiding decisions in the field of accident prevention and intervention using MobileNet and Transformer learning.