A Performance Evaluation of Time-Series Mobility Prediction for Connected Vehicular Networks

Abstract

The role of connected vehicular networks has become vital for the future of smart and modern cities, as it can be envisioned as a stand alone connected network or a bridge between various networks, and end-users. Vehicular networks design provides safety, traffic control management, and cruise control services, among others. Many applications and protocols such as mobility management, service discovery, and routing were proposed over the past years to elevate the performance of vehicular networks communication and connectivity. In that context, understanding the vehicles' mobility characteristic, behavior, and pattern would assist those applications to become user-centric, adaptive, and proactive based on the dynamics of vehicles' projections. In this paper, we evaluate the performance of time-series forecasting techniques for vehicular mobility prediction. We first describe and extract vehicles mobility features into time-series sequences and discuss several machine learning techniques. Then, we compare the performance of recent time-series ML-based techniques, LSTM, and GRNN in terms of complexity and accuracy.