Utilizing GIS and Machine Learning for Traffic Accident Prediction in Urban Environment

Abstract

Traffic accident prediction is crucial to preventive measures against accidents and effective traffic management. Identifying hotspots can facilitate the selection of the most critical survey points to note the contributing features. In this research, an effort has been made to identify hotspots and predict traffic accident occurrences in an urban area. Accident data was obtained from the Rescue 1122 Emergency Services of Faisalabad, and hotspots were identified using Moran’s I in ArcGIS. Results showed that most hotspots were located around the General Transport Stand (GTS) area due to the maximum number of road users. The temporal investigations showed that the accident occurrence was significant from 1 to 2 p.m. The identified hotspots were further investigated by conducting a field survey. Essential features such as road geometric features, road furniture, and traffic data were used for developing Machine Learning Algorithms for accident prediction. Using Computer Vision, traffic data was extracted from recorded videos. Random forest, linear regression, and Decision tree algorithms were developed using Python in the Jupyter Notebook environment. The decision tree algorithm showed a maximum accuracy of 84.4%. The analysis of contributing factors revealed that road measurements had the maximum effect on accident occurrence.