Improving Accuracy of Network Bandwidth Prediction Using Multi-Channel Gaussian Process Regression

In this paper, we present an effective model for predicting network bandwidth using multi-channel time series data. Traditional methods such as stochastic ARIMA and deep learning-based LSTM are common; however, recent studies emphasize the advantages of Gaussian Processes (GP) in handling non-linear data structures, requiring less data, and quickly adapting to changes in signal characteristics. Our model leverages a multi-Output Gaussian Process (MOGP), improving prediction accuracy by exploiting inter-channel correlations. We introduce an approach that generates new signals from the original signals by summing adjacent signals, enhancing the input signal set. Our results show enhancements over traditional methods, demonstrated through experiments using a dataset of multiple signal channels processed and analyzed for correlations and future trend predictions. Predictive performance is quantified through the Root Mean Squared Error (RMSE) metrics compared to true future values, showcasing our model's capability to effectively predict network bandwidth with high accuracy and reliability.