Publication

Abstract : Road accidents are increasing day by day. In India, every year nearly 33% of road accidents take place in national highways and 23% in state highways. This paper focuses on predicting the road accidents in national and state highways in India using machine learning algorithms. The data is curated using annual reports published on road accidents in India by ministry of road transport & highways also, the reports on the compensation to the victims by motor vehicle accident fund trust. Road accident details are captured from the year 1990-2022. This covers the information about monthwise, state-wise accidents due to all types of vehicles, road environment, type of traffic control, vehicle condition, weather condition, rural/urban areas and so on. The data contains information about total number vehicles registered, road length, number of accidents in a year, number of people killed and the number of people injured. The road accidents and the number of vehicles causing accidents are predicted using machine learning algorithms viz., Linear Regression, Ridge Regression, Decision Tree, K-Nearest Neighbour (KNN) and Multilayer Perceptron algorithms. Multilayer perceptron model outperformed all other prediction models with highest R2-Score of 0.93278 and Mean Squared Error (MSE) of 0.277503 and Root Mean Squared Error (RMSE) of 0.387880 for five-fold cross validation. Prediction helps in identifying the critical regions/roads where the number of accidents are more. This helps in implementing accident preventive measures like restricting speed limit in such highways.