A Deep Learning Approach for a QoS Prediction System in Cellular Networks

In this research, we address the critical problem of predicting Quality of Service (QoS) in cellular networks, a key factor in ensuring a seamless user experience in a world of ubiquitous computing and giant data networks. Using deep learning techniques, we predict future complex lags based on time series data in three simulated drift scenarios. Created in OMNET++, Simu5G, Veins, and Sumo, these scenarios include adding a base station, building a new tower, and modifying mobility templates for mobile clients. Each scenario includes more than 2 million samples. Our study uses the N-BEATS model, a deep learning architecture specialized in time-series forecasting that can identify unique patterns and trends to predict future values. The implementation was done with PyTorch, including data preparation, creation of N-BEATS model architecture, configuration of loss functions and optimizers, and training of the model for prediction. Our goal is to present an efficient, scalable and interpretable solution to predict cellular QoS changes in dynamic environments, thereby increasing network performance and user satisfaction. Our work aims to make an insightful contribution to the field of QoS prediction in cells by demonstrating the potential of N-BEATS. Evaluation metrics such as MSE, RMSE and MAE are used for error estimation.