Deep Learning Approaches for Traffic Prediction Forecasting in Multi-Level Cloud Architecture for Autonomous Vehicle Services

Abstract

Autonomous vehicle (AV) is one of the most new use cases and a technology for fifth-generation (5G) and next-generation mobile networks in numerous applications., the use of AVs has exponentially worldwide due to the rapidly growing awareness and use of artificial intelligence (AI) methods in various fields. Predicting data flows is essential for AVs to improve data transmission and decrease delays through more efficient use of appropriate capabilities, monitoring, management, and control of the traffic system. This paper proposes a deep learning approach (DL) with the bidirectional long-short-term memory model (BI-LSTM) for predicting the traffic rates of AVs with multi-cloud services. In terms of prediction accuracy, a comparison is conducted between the BI-LSTM and the unidirectional LSTM based on the number of batch sizes used. The prediction accuracy is computed using the root mean square error (RMSE), mean absolute percentage error (MAPE), coefficient of determination (R2), and processing time. The results show that the prediction accuracy with BI-LSTM outperforms the LSTM model. Besides, the prediction accuracy using 8 batch sizes outperforms the competitors and offers outstanding performance.