1.35 million people die every year globally in road accidents (much worse than the recent COVID-19 pandemic). Road accidents have become a global menace and are a significant public health concern. Can Artificial Intelligence technology be an enabler and mitigate the road accident risk? This research is published titled “Predicting and explaining the severity of road accident using artificial intelligence techniques, SHAP and feature analysis” and published in the International Journal of Crashworthiness, Taylor and Francis Group, a SCOPUS journal with an impact factor of around 3.

The study envisages an AI-enabled accident management system that can predict, send automatic alerts and communicate efficiently within various channels. This can save the lives of accident victims and can help manage traffic and debris well. Machine learning-based scalable and accurate prediction of severity, among others, is pivotal to a reliable accident management system. By addressing the issue of scalable and accurate AI model predictions, the current study assumes significance and aims to motivate further research in this area.

Abstract of the paper:

The accurate prediction of severity is crucial to mitigating road accidents’ public health risks and improving traffic safety. Thus, it has become an active area of research in recent years. However, studies in certain regions such as South Asia and Sub-Saharan Africa are comparatively less, though they share a large proportion of global accidental deaths and injuries. Secondly, traditional methods are predominantly used, although artificial intelligence (AI) techniques have been experimented with in recent years to predict accident severity. Lastly, a limited number of AI studies have carried out any feature analysis to explain model predictions. In this study, we aim to contribute in four ways: (i) we propose an analytical review of the literature to gauge the interest and scope of existing studies and identify the direction for further research, (ii) a mixture of old and relatively new AI techniques such as gradient boosting machine (GBM), extreme gradient boosting (XGB), random forest (RDF) and support vector machine (SVM) is applied to road accident data of India, (iii) we employ Shapley additive explanations (SHAP) for interpretation of AI model predictions, and (iv) we propose an AI-enabled accident management system which can save millions of lives subject to proper implementation. The findings suggest that AI models are greatly capable of predicting the accident severity. Among AI models used, GBM attains the best test accuracy for predicting severity with precision, recall, and f1 of 0.92, 0.85, and 0.88 for the injured class and 0.86, 0.93, and 0.89 for the killed class, respectively. On computation time required, GBM is better than RDF and SVM, however, it is slightly behind XGB. Out of five broad feature categories with 19 features, vehicle type is found as the most prominent feature category to influence severity predictions. Specifically, commercial vehicles, excess speed, national highways, overloading, and pedestrian fault are important factors responsible for accidental road killings as per the SHAP values.

The paper is authored by the four Alumnus of the School of Economics, University of Hyderabad.

They are Dr. Chakradhar Panda, Director of Data Science, Publicis Sapient, Bengaluru, Karnataka; Dr. Alok Kumar Mishra, Associate Professor, School of Economics, University of Hyderabad; Dr. Aruna Kumar Dash, Department of Economics, IBS Hyderabad, IFHE University, Hyderabad, and Mr. Hedaytullah Nawab, Uber, Hyderabad, Telangana, India.


Dr. Alok K Mishra               Dr. Chakradhar Panda                     Dr. Aruna Kumar Dash                 Mr. Hedaytullah Nawab

The article can be referred to at the link below: