Performance Analysis of QOS Parameters and Churn Prediction Model Development for MNOs in Nigeria

Abstract

This paper presents the results and analysis of research about subscriber satisfaction and their propensity to churn in the Nigerian GSM Telecommunication Industry. The research built a model using three different algorithms, viz: Principal Component Analysis (PCA), KMeans Clustering for unsupervised learning of dataset, and the Logistic Regression (LR) Model. The datasets used were performance metrics of the Quality of Service (QoS) of four major Mobile Network Operators (MNOs) in Nigeria; Airtel, 9mobile, Globacom, and MTN. The QoS were analysed using a built model. The parameters of the QoS analysed and from which a model was built were the Call Set-up Success Rate (CSSR), the Drop Call Rate (DCR), the Standalone Dedicated Channel Congestion Control Rate (SDCCH), and the Traffic Congestion Control Rate (TCCH). The PCA helped to simplify the datasets such that the trends and patterns are retained and thereafter, the KMeans Clustering was applied on the simplified dataset and was tagged with segregating the Network Predictors which indicated the MNOs with satisfactory results and those likely to lose subscribers. The LR model was used in a supervised learning mode for unlabeled datasets using clusters as labels which finally resulted in a prediction model with a predicting accuracy of 95%.