When and where Proactively predicting traffic accident in South Africa: our machine learning competition winning approach

Abstract

South Africa (SA) records high mortality originating from traffic accident annually making the country to be ranked highly among nations with the highest traffic mortality globally. There is seemingly no study that has attempted to forecast when and where next accident will occur in SA. This study aims to use machine learning method to predict traffic accident in SA for every hour ranging between 1 January and 31 March 2019 at a segment ID. We obtained details of accidents that occurred in Cape Town, SA between 2016 and 2019 SANRAL, Uber Movement and Cape Town FMS via Zindi competition platform. This research adopted Catboost and LightGBM models to predict the traffic incident occurrence. Our model shows a F1 score of 0.11. The results of this research will aid prediction of accident occurrence at a particular road segment hourly.