Comparison of supervised machine learning algorithms for road traffic crash prediction models in Rwanda

Abstract

As the country developed, and there was a rapid growth of various modes of transport, as well as the occurrence of road traffic crashes. Rwanda also faced the same challenges of road traffic crash severity, in which every year the number of fatalities increased progressively. To overcome these challenges, the study has focused on comparing the classification performance of eight supervised machine learning algorithms in order to visualize which is the best to predict crash severity and identify the potential crash-influential factors in Rwanda. The quantitative datasets of road traffic crashes, registered vehicles, and AADT have been used from 2010 to 2022. The ML algorithms, including LR, SVM, NB, K-NN, RF, DT, LBR, and J48, have been employed. The model results indicated that five algorithms, including RF, DT, J48, LBR, and K-NN classifiers, have shown better accuracy, greater than 80%. The RF had the highest ability to predict the crash severity in Rwanda, with an accuracy greater than 97%. The most identified influential factors were AADT, registered vehicles, causes of crashes, and vehicles involved. The model results can be applied to provide useful information to road safety decision-makers during the planning and design of road infrastructure.