Abstract: Real-time crash risk analyses play a vital role in Active Traffic Management Systems (ATMS) by identifying the hazardous traffic conditions that potentially precede crash occurrences within a very short time. Currently, the recent advancements in traffic sensing and detection technologies have had a tremendous impact on real-time crash risk safety analysis. However, there is a lack of prior studies that attempted to examine the relationships between crash occurrence and real-time traffic data collected from floating cars on expressways. Moreover, several researchers mostly developed real-time crash prediction models with resampled balanced datasets which may be inadequate to the large continuous real-time traffic data environment. Therefore, in this study, a comprehensive imbalanced classification algorithm, Adaptive Boosting Algorithm for Convolutional Neural Networks (AdaBoost-CNN), has been first time introduced to build a practical real-time crash prediction model. This study primarily aims to: (1) investigate the feasibility of using Floating Car Data (FCD) to predict the real-time crash risk on expressways; and (2) explore the efficiency of AdaBoost-CNN algorithm to solve the imbalanced data classification problem. Two models are compared to the proposed AdaBoost-CNN. First, AdaBoost with CNN base classifiers is compared to the proposed model to investigate the influence of transfer learning on prediction accuracy. Second, One-Dimensional Conventional Neural Network is designed with balanced data to examine the capability of AdaBoost-CNN to handle the imbalanced data issue. Experiments demonstrate the high accuracy of AdaBoost-CNN in predicting crash and non-crash cases in the context of sensitivity, false alarm rate, and Area under Curve scores.
Ahmad Yahia, Xuesong Wang*, Tonggen Wang. Utilizing the Imbalanced Classification Algorithm for Real-time Risk Analysis on Expressway Using Floating Car Data. Transportation Research Board 101st Annual Meeting, Washington D.C., USA, 2022. 1.9-13.