Improving road safety in smart cities using machine learning techniques

Road safety is a serious concern in urban environments and smart cities, involving the well-being of all road users, with a special emphasis on those considered vulnerable such as pedestrians, bicyclists, and motorcyclists among others. Being a difficult challenge, it becomes evident that the traditional methods of road safety are not sufficient to address these challenges and needs a more Intelligent Transportation Systems (ITS). This study aims to analyze traffic accident data, identify factors behind severe injuries, and develop predictive solutions using Machine Learning (ML). For this study, data were collected from the Public Safety Data Portal in Toronto and the District Emergency Office in Rawalpindi, Pakistan (RTA). The data were preprocessed by applying different data preprocessing methods using Python. A number of popular ML and Deep Learning (DL) models have been trained and tested on the preprocessed datasets for traffic accident analysis. Results showed that the XGBoost and Random Forest exhibited excellent performance with an accuracy of 74% on KSI dataset without under-sample and hyperparameter tuning methods. Random Forest achieved high accuracy 99% after applying the Grid Search method of hyperparameter methods and undersample technique Moreover, the current study has utilized the Association Rule Mining technique to determined the underlying hidden factors from both dataset that lead to collisions and fatal or major injuries. The extracted rules revealed that certain factors, such as over speeding, aggressive driving, pedestrian collisions, disobeying traffic rules, lost control, driver’s inattentiveness and the absence of traffic controls at major roads and intersections are associated with a higher risk of fatal or major injuries. Furthermore, the study provided a comprehensive comparison of factors contributing to severe collisions in Pakistan and Canada. It found that there is a need for targeted enforcement in both countries including stricter licensing and education initiatives for young drivers about responsible driving behaviors. The proposed framework can be deployed in a real environment for improving road safety in urban environments and Smart cites.