Prediction of user throughput from Network Parameters of LTE Network using machine learning

The number of subscribers is increasing with every passing day, and each user requires a diverse type of service. This leads to an increased load on the network, thereby degrading the quality of service. It becomes a challenge for the operators to adapt to such changes rapidly without the need to deploy new hardware in the network. One of the major hindrances faced by the operators is the estimation of throughput. If the throughput could be predicted before a connection has been established between the Base Transceiver Station (BTS) and the user, the network performance could be enhanced. This would further help in minimal wastage of resources and maximum utilization of resources, and thus benefit the operators as well as the customers in a significant manner. This work aims to provide a potential method to predict the uplink and downlink throughput from the user perspective using various machine learning algorithms. Maximum user throughput has been calculated using mathematical equations. A dataset of 135 features of Nokia Network Pvt. Ltd. containing the information of nearly 50,000 base stations has been used in this work. The dataset was divided into four categories based on the average uplink and downlink throughput values where these average values were calculated from dataset, upon which three machine learning algorithms viz. Support Vector Machine (SVM), Naïve Bayes and K-Nearest Neighbours (K-NN) were implemented for the prediction of the user throughput category. The performance of the models is compared using confusion matrix and classification report. The maximum accuracy of 96.17% for downlink and 96.10% for uplink was achieved using SVM.