An Ensembled Approach to Time Series Prediction for Vehicle Communication

Abstract

Machine Learning (ML) for accurately modeling and predicting the wireless channel for various application especially in vehicle is essential for increased safety. In some scenarios we need long distance communication without increasing the congestion factor for the V2V communication. Beam forming is a method of increasing the range but choosing the beam is a critical where machine learning shall be used to predict which beam to choose and predict the power level. In this paper, we design an ensemble machine learning of time series based on three machine learning concepts which are LSTM, ARIMA and Random Forest. Each machine learning concept as a standalone has a unique advantage and our ensemble approach provides the better accuracy of prediction. The ensemble approach is set for multi-step prediction where as one step prediction is more of following rather than predicting. We evaluate our ensemble approach with the standalone algorithm.