Traffic Crash Countermeasure Recommendations Using Deep Neural Network: A Decision Support Tool for Traffic Safety Engineers

초록

Various measures have been implemented to reduce the number of traffic crashes, which can cause significant social and economic damage. Several studies have identified the factors that cause traffic crashes and recommended effective countermeasures for specific types of traffic crashes. However, only a few studies support decision making by comprehensively presenting the causes and countermeasures of crashes. Various policy making stakeholders are emphasizing the need to establish a traffic safety management knowledge database that presents comprehensive information on crash causes and countermeasures. Therefore, this study aims to support reasonable decision making by recommending appropriate countermeasures based on crash types using a deep neural network (DNN). The algorithm automatically recommends countermeasures suitable to each type of traffic crash. The DNN-based multilabel classification label model learns the recommended countermeasure use-case results of each matched traffic crash use-case using logic. The algorithm is validated by applying the K-fold cross-validation. A performance evaluation of the model reveals that its accuracy reaches 93%, and other evaluation indicators, such as precision, recall and f1-score exhibit excellent results. An algorithm that automatically recommends traffic crash countermeasures can help policymakers make rational decisions based on the established database. An algorithm that automatically recommends traffic crash countermeasures can help policymakers make rational decisions based on the established database. The algorithm supports policymakers, such as public officials in transportation departments and government agencies, responsible for implementing traffic safety measures. It overcomes the limitations of fragmented reports by providing data-driven recommendations, enabling efficient resource use and implementation of optimal safety strategies. © 2013 IEEE.

키워드