Rear-End Vehicles Collision Detection in Internet of Vehicles Using Explainable AI and Machine Learning

Abstract

In recent days, with the high rise in population, increase in possession of multiple vehicles in a family which creates heavy traffic on roadways, vehicle collision is one of the principal causes of death worldwide due to road accidents. Internet of vehicles (IoV) has become a salient innovation of intelligent transportation system (ITS), which is attracting attentions of researchers and scientists recently to assist the drivers to have a safe driving experience. With the tremendous opportunities in today’s autonomous world, explainable artificial intelligence playing a remarkable character to address the censorious facet of the IoV with a detailed explanations while detecting rear end collision of vehicles in IoV to have quality of services and overall safety. Previous research applied machine learning and deep learning algorithms for the detection of rear end collisions in IoV but have limitations in not using explainable artificial intelligence (xAI) techniques which would dramatically enhance the interpretability and performance acceptance of the IoV overall system. This paper uses LIME and Captum’s DeepLift attribution xAI methods to understand the inner details of the model prediction in IoV scenario. The IoV datasets are collected from Kaggle for the proposed research. The performance of the proposed xAI based machine learning approaches using several Boosting algorithms, Random Forest and Gated Recurrent Neural Network approaches, for rear end collision detection in IoV are measured with extensive simulations in this research. The experimental results show that the proposed work is accurate in detecting the rear end collision of vehicles in IoV environment with detailed explanations which shows its effectiveness, transparency and trustworthiness in improved public safety in IoV environments