Road Crashes Analysis and Prediction using Gradient Boosted and Random Forest Trees

Abstract

People lose their lives every day due to road traffic crashes. The problem is so humongous globally that the World Health Organization, in its Sustainable Development Agenda 2030, is inviting the coordinates efforts across nations towards it and aspiring to cut down the deaths and injuries to half. Taking a clue from that, the proposed work is undertaken to build machine learning-based models for analyzing the crash data, identifying the important risk factors, and predict the injury severity of drivers. The proposed work studied and analyzed several factors of road accidents to create an accurate and interpretable model that predicts the occurrence and severity of car accidents by investigating crash causal factors and crash severity factors. In the proposed work, we employed three machine learning algorithms to vis-à-vis Decision Tree, Random Forest, and Gradient Boosted tree on Statewide Vehicle Crashes Dataset provided by Maryland State Police. The gradient boosted-based model reported the highest prediction accuracy and provided the most influencing factors in the predictive model. The findings showed that disregarding traffic signals and stop signs, road design problems, poor visibility, and bad weather conditions are the most important variables in the predictive road traffic crash model. Using the identified risk factors is crucial in establishing actions that may reduce the risks related to those factors.