Video

Abstract

Traffic accidents have emerged as a serious global concern, resulting in daily casualties and prompting authorities to prioritize accident prevention measures. This study presents an accident prediction system that alerts drivers of potential accidents by analyzing multiple attributes which are the potential courses of accidents. While previous studies have predominantly focused on analyzing geographical factors, predicting accident frequencies, and assessing accident risks, this study aims to develop two systems—an advanced route recommendation system and a real-time accident prediction system—to enhance road safety. Moreover, existing systems often predict post hoc accident occurrences or have limited geographical coverage. The advanced route recommendation system is designed to assist users in planning their journeys by providing them with the safest routes in advance. Through a website interface, users can log in and receive personalized recommendations on the accident prone areas on their path, based on factors such as historical accident data, road conditions, traffic patterns and weather conditions. This system aims to empower individuals to make informed decisions and reduce the likelihood of accidents during their trips in advance. The real-time accident prediction system aims to provide drivers with up-to-date information on potential accidents along their routes. By utilizing GPS coordinates and retrieving live data, including weather conditions and real-time accident reports, the system predicts accident-prone areas in real-time. Drivers receive these predictions through a mobile application as an audio message, enabling them to make timely adjustments to their routes and avoid hazardous situations. Additionally, static predictions are displayed on a website, featuring markers indicating accident-prone areas. The research utilizes an extensive dataset of “US Accident Dataset” spanning across all 49 states of the USA. Results demonstrate that the Random Forest classifier achieves an impressive 91.5% accuracy in predicting accident severity, surpassing previous studies. Furthermore, this paper conducts Exploratory Data Analysis, unveiling intriguing patterns in the dataset regarding accident occurrences.

Citation

Acknowledgements

This research was supported by the Science and Technology Human Resource Development Project, Ministry of Education, Sri Lanka, funded by the Asian Development Bank (Grant No. STHRD/CRG/R1/SJ/06).

The website template was borrowed from Michaël Gharbi and Ref-NeRF.