A methodology for prioritizing safety indicators using individual vehicle trajectory data

Abstract

A methodology for assessing crash risk using vehicle drivingtrajectories based on data mining techniques was developedin this study. A variety of safety indicators reflecting the char-acteristics of traffic and road geometric conditions were eval-uated in terms of their capability of capturing hazardoustraffic flow. Comprehensive data preparation was conductedby matching driving trajectory data obtained from in-vehicledigital tachograph devices and crash data to classify and ana-lyze hazardous and normal traffic flows. The random forestapproach was adopted to quantify the importance of safetyindicators. The crash risks were evaluated using the logisticregression model and multivariate adaptive regression splinesmodel based on the set of safety indicators with high import-ance. The results show that the dangerous driving events rateand driving volatility indicators were found to be particularlysignificant in identifying hazardous conditions. The multivari-ate adaptive regression splines model showed better perform-ance and a classification accuracy of 86% was achieved. Theproposed methodology will be useful for deriving effectivecountermeasures to prevent crashes, which is the backbone ofproactive traffic safety management.