Prediction of Road Accident and Severity of Bangladesh Applying Machine Learning Techniques

Road accidents in Bangladesh have become widespread nowadays. This not only damage our economies but also affect many families as well. Earlier researchers in Bangladesh had suggested a few approaches to machine learning and computer vision, but they had some limitations. Prior work on this issue mainly focused on either the possibility of an accident or the degree of severity. Some works showed low accuracy due to deficiency in record which is a major problem. They either predicted accident with one or two specific factors or only found the factors related to the accident. Through this paper, we proposed a multiclass model in which we combined both the prediction of accidents and their corresponding severity to develop a better model to avoid road collisions. We also merged five accident related casualties to properly interpret the nature of the accident. Analyzing sixty factors of five causalities we used different machine learning algorithms for prediction. Among them Decision Tree, Random Forest, Multilayer Perceptron and Categorical Naive Bayes showed acceptable result, but the best outcome obtained with Decision Tree. This algorithm obtained a strong accuracy of 99.77% for accident prediction and 99.80% for severity prediction With an F1 score of 98.68% and 99.80%.