LTE QoS Parameters Prediction Using Multivariate Linear Regression Algorithm

Quality of service indicators monitoring and prediction are fundamental prerequisites of an LTE broadband wireless network for satisfactory service delivery of evolving Internet applications to customers. The mean user throughput and the delay are key performance indicators(KPIs) that have to be monitored against the evolution of the network in terms of radio quality and total traffic. This method allows optimization engineers to have accurate idea about the current quality of service provided to the customers and take appropriate actions to manage the network resources. Today’s popular mobile Internet applications, such as gaming, voice service, streaming and social networking applications, have diverse traffic models and, consequently, different QoS requirements. The engineers should have an idea about the current user throughput and the application delays that characterizes the network and should be able to predict the evolution of such indicators in the coming months to enrich the marketing team knowledge about the services and the offers that can be suggested to the subscribers. In this paper, we use a learning data set that contains the daily evolution of the four KPIs (mean user throughput, average delay, total traffic carried over the network and the average reported channel quality indicator (CQI)) for a wireless operator during six months and we implement the machine learning multivariate linear regression technique to prove that the mean user throughput and the average delay can be estimated by a linear function of the total traffic carried over the network and the average reported CQI using gradient descent algorithm. Then we evaluate the model that we got by calculating the residuals. Finally, we discuss several case studies to show the potential of the optimum linear function in describing the network QoS evolution pattern and the tuning of existing features.