Explainable Customer Churn Prediction in Telecom Using Ensemble Learning and SHAP Analysis

Abstract

There is a fierce competition amongst telecom service providers in many geographies. Every business is striving to minimize the churn of its customers. The churn prediction approach helps to identify the potential high-risk customers who are planning to discontinue the services. This work has analyzed this problem with the aid of multiple machine learning algorithms to assess what can be the most effective techniques to predict the churn. Before getting to the machine learning algorithms, the data was cleansed for missing values, impute where possible, normalize to the same scale and encode the non-numerical values using techniques such as one-hot and frequency encoding. This work has analyzed all the leading classification algorithms like Random Forests, Decision Tree, Artificial neural network, Extreme Gradient Boosting and a few others. Random Forest keeps performing the best on the available data set, but this work has been able to improve the accuracy of several other methods as well by proper pre-processing of the data and hyperparameter tuning. To make the functioning of these methods transparent to the business managers - this work has used SHAP as the explainable AI technique. This also corroborates the initial findings from the exploratory data analysis. This work indicates how critical it is for the telecom business to utilize these methods to predict and preempt the chances of churn which are denting their business prospects.