Customer Churn Prediction in Mobile Networks using Logistic Regression and Multilayer Perceptron(MLP)

Abstract

Customer relationship marketing is important since it provides a long standing relationship between the customer and the organization. Churn obstructs the growth of profitable customers and it is the biggest challenge to sustain a telecommunication network. We propose two models which predicts customer churn with a high degree of accuracy. Our first model is a logistic regression model which is a non-linear classifier with sigmoid as its activation function. The accuracy of the model is heightened by regularizing it with the regularizing parameter set to 0.01 and this gives an accuracy of 87.52% on our test dataset. Our second model is a full fledged Multilayer Perceptron(MLP) Neural Network with a normalized input feature vector which is stacked with three hidden layers and employs binary cross entropy as the loss function with a learning rate of 0.01. This model is split into a test-train set and achieves an accuracy of 94.19%. Using this predictive model the organization can conduct marketing research and study the needs of the particular customer in detail. Using that data they can produce goods according to the customer needs before the customer demands and present it to them. This helps to create brand loyalty which in turn leads to a sustainable network.