PhD Seminar: A temporospatial context-aware vehicular crash-prediction model

Candidate: Pouya Mehrannia

Title: A temporospatial context-aware vehicular crash-prediction model

Date: October 17, 2018

Time: 3:00 PM

Place: EIT 3145

Supervisor(s): Basir, Otman - Moshiri, Behzad (Adjunct)

Abstract:

With the demand for more vehicles increasing, road safety is becoming a growing issue. Much of the empirical research on road safety and accident prediction utilize statistical models that capture limited aspects or contributors. While some risk factors contribute more frequently in the occurrence of a road accident, the importance of driver behavior, temporospatial factors, and real-time traffic dynamics have been underestimated. The objective of this study is to establish a crash prediction model (CPM) based on historical accident data, driver behavior, and environmental and location data. The proposed CPM predicts traffic accidents by incorporating machine learning and data analytics techniques to identify driving patterns and other risk factors associated with potential vehicle crashes. These techniques include clustering, association rule mining, data fusion, and Bayesian networks. The results obtained from the preliminary work performed based on the national collision database of Canada show that valuable insights can be gleaned from the collision databases. Association rule mining uncovers the underlying rules in the collision databases and data segmentation eliminates the effect of hidden variables. These rules combined with a real-time processing of relevant crash contributors can pave the way for predicting crashes and their severity in real-time. A data fusion framework is proposed to enhance the model prediction accuracy by integrating pieces of evidence discovered from different datasets. It is conceivable that such prediction model will provide drivers crash likelihood warnings so as to prevent accidents in real-time.