A Framework for QoS Data Collection and Throughput Prediction in 5G Networks Using Deep Learning Models

This research aims to develop an artificial intelligence-based model for predicting throughput in 5G networks. Additionally, this study also aims to develop a Quality of Service (QoS) data collection tool for 5G networks in a controlled environment to enable data acquisition from the 5G network platform. Although 5G networks are expected to provide high throughput, low latency, high availability, and reliability by 3GPP standards, there has been no large-scale field testing reported in the literature regarding QoS measurements in direct public settings. Therefore, in order to fill the knowledge gap related to the understanding and the prediction of service quality in the context of 5G networks, this research develops a QoS data collection tool for 5G networks and applies multivariate time series analysis and machine learning techniques to predict QoS metrics in 5G networks. Several innovations introduced include the use of Non-Standalone (NSA) 5G networks to handle various types of files, new contributions to the development of research infrastructure through the creation of a QoS data collection tool in a controlled environment, new considerations related to changes in network behavior under diverse user conditions, and the application of multivariate time series analysis techniques and deep learning models to predict 5G network throughput. This research employs a multivariate time series analysis approach to predict QoS metrics in 5G networks. The machine learning techniques to develop prediction models include RNN, LSTM, GRU, and LSTM-ED. The collection of 5G network QoS data is carried out by using a specifically developed data collection tool. The collected data for four deep-learning models have been processed and adapted to predict network throughput. Model performance evaluation uses MSE, RMSE, MAE, and WMAPE metrics. The outcome of this research is the development of a QoS data collection tool for 5G networks in a controlled environment, multivariate time series analysis, and throughput prediction by using deep learning models. The study indicates that the LSTM-ED model provides the highest level of accuracy in predicting 5G network throughput, with an accuracy rate of 91.21%. The research also demonstrates that changes in network behavior under various user conditions can influence the quality of service in 5G networks.