Efficient Collision Risk Prediction Model for Autonomous Vehicle Using Novel Optimized LSTM Based Deep Learning Framework

Rear-end collision prediction has attracted more and more attention in order to increase safety in Autonomous Vehicles (AVs). It is critical to develop effective warning systems for autonomous vehicles because rear-end collisions are one of the major contributors to traffic accidents. The current research has been done to foresee collisions. We suggest learning-based approaches to address this complex problem, which the conventional approaches find challenging. However, due to non-optimized parameters in the Long Short Term Memory (LSTM), the prediction performance is degraded. To overcome this issue, Novel, optimized LSTM-based deep learning framework was proposed to enhance Crash Risk Prediction System effectiveness. This framework uses an Enhanced LSTM (EnLSTM) to improve the performance of Crash Risk Prediction system while extracting efficient features using a Convolutional Neural Network (CNN)-based feature extraction mechanism. Improved Grasshopper optimization Algorithm (IGOA) has been developed for hyper parameters optimization in the LSTM model. The proposed algorithm is assessed on the basis of metrics from confusion matrix. The experiment reveals that the Novel Optimized LSTM is superior to the current models based on precision, sensitivity, specificity, false and missed alarm rate. This efficient deep learning model may be applied in autonomous vehicles and accurately warns the driver when a collision is imminent.