Extracting Traffic Safety Knowledge from Historical Accident Data

Abstract

This paper presents a method and a tool for analyzing historical traffic accident records using data mining techniques for the extraction of valuable knowledge for traffic safety management. The knowledge is dis-tilled using spatio-temporal analysis of historical accidents records. Raw accident data, obtained from Police records, underwent pre-processing and subsequently integrated with secondary traffic-flow data from a mesoscopic simulation. Clustering analysis was performed with self-organizing maps (SOM) to identify accident black spots on the road network and visualizethis on a map. Distilled knowledge is used to develop a prototype mobile application to warn drivers of accident risk in real time