CHURN PREDICTION OF CUSTOMER IN TELECOMMUNICATION AND E-COMMERCE INDUSTRY USING MACHINE LEARNING

Abstract

Customer behaviour can be represented in many ways. The customer's behaviour is different in different situations will give his idea of customer behaviour. From a general perspective, the behaviour of the customer, or rather any person in any area, is taken to be random. When observed keenly, it is often seen that the future behaviour of a person can depend on various factors of the present situation as well as the behaviour in past situations. This research constitutes the prediction of customer churn, i.e. whether the customer will terminate purchasing from the buyer or not, which depends on various factors. We have worked on two types of customer data. First, that is dependent on the present factors which do not affect the past or future purchases. Second, a time series data which gives us an idea of how the future purchases can be related to the purchases in the past. Logistic Regression, Random Forest Classifier, Artificial neural network, and Recurrent Neural Network has been implemented to discover the correlations of the churn with various factors and classify the customer churn efficiently. The comparison of algorithms indicates that the results of Logistic Regression were slightly better for the first Dataset. The Recurrent Neural Network model, which was applied to the time-series dataset, also gave better results.