A Decision support system to predict highway accident alerts in Sri Lanka

Abstract

Road traffic and rapidly increasing road accidents become a vast problem not only for Sri Lankans but for all human beings who are living in this planet. According to the Sri Lanka police highway reports, there have been a number of deaths recorded due to fatal accidents in highways and also affected significant amount of people for non-fatal accidents as well. Several key factors vastly contributed directly to cause an accident, such as environmental factors, human factors and road condition factors., etc. In this research, machine learning techniques and methods have been applied to the southern highway accident records retrieved by Police Highway Control Center for the period of 2015 to June 2019 in order to establish a model which enables to forecast reason for the accidents that will be occurring in the future. The python scikit-learn package has been used on top of the anaconda framework to discover hidden patterns of data with the help of decision tree, support vector machine and logistic regression powered by a one-vs-rest classifier. The two well-known ensembles; random forest and gradient boost classifiers have also performed in this dataset in order to enhance the accuracy. The performance of each classifier has compared critically based on their results. The results obtained by performing various experiments show that the ensemble’s work well when compare to other classifiers.