Prediction of Risk Generated by Different Driving Patterns and Their Conflict Redistribution

Abstract

In this paper, risk level correlation and classification using belief functions based on driving activities' data about sharp transient maneuvering tasks, legal speed exceeding, and average speed ensuing with the human being who is controlling the technical system, i.e., the car, is presented. A dataset is constituted by time stamped and geographically referenced driving maneuver information, which is exceptionally reported when an acceleration exceeds the given threshold in both longitudinal and lateral direction. Risk level is labeled in terms of the change in total vehicle collision property damage cost for the analyzed time period, and long term driving activities' magnitude and frequency is divided into two equal time periods, which are used to predict risk levels. Redistribution of conflicts generated by driving activities is used to predict the future risk level of involvement in an accident. Using fuzzy approach a microaveraged F-measure 90.98% is achieved by Dubois-Parade and PCR6 conflict redistribution methods.