Traffic crashes prediction using machine learning models, case study: Rwanda

Abstract

Predicting traffic crashes has become a significant emerging challenge yet Road traffic accidents in Rwanda accounts for 5000 crashes annually on average, claiming 54% of the lives of pedestrians and cyclists. And recent research attempts to tackle this area are limited to summary statistics. This study aims to build a predictive machine learning model so that road traffic accidents can be predicted and policymakers and road traffic users can make informed decisions. Based on the historical road traffic accidents dataset in Rwanda and python programming language, two supervised predictive machine learning models were trained and test on the sample accidents records for six years registered by RNP/ Traffic department across the country. The first model predicted the number of accidents. And the second model classified the accidents based on injury severity categories. The model results indicate the regression model correctly predicts the total number of accidents 100%. The model prediction results match the actual accidents records. The random forest model classifies accidents injury severity at the rate of 91%. Both models are recommended for use as a key to prediction to better prevent road traffic accidents. Moreover, it is recommended that the road traffic accidents database keep daily accidents records to improve the prediction. Further researchers may focus on building automated systems that integrated information from driver’s behaviours, road, weather, vehicle data instantly.