Customer Churn Analysis Using Machine Learning

Abstract

Many companies lack awareness about the different kinds of customer deviations that exist in today’s world. There can be many reasons that factor the churn rate of a company, ranging from the success of a product, reputation of the brand, extra services, accessibility, price range and many others. It’s usually very tedious to shortlist a particular reason that is causing a higher churn rate than the others manually. Recognizing this problem, this paper answers some of the churn analysis questions through the development of an efficient churn analysis machine learning based model that performs various functions. The proposed work is broken down into two phases. First being, data analysis followed by churn prediction. For data analysis, multiple graphs are plotted with different features to gain interesting insights on the shape and nature of the company’s churn rate and to narrow down on which combination of features might be more heavily correlated with the predictor variable ‘churn’. For churn prediction, a classification model was built that comprised of six algorithms. Further on, cross validation and hyperparameter tuning was performed on all the models. An ensemble model was also built to increase model accuracy and finally, performance evaluation was done to check the best built model. Ultimately, the model giving the best results in the performance evaluation phase is chosen to be used for the end-to-end model use. In the proposed work, XG Boost Classifier proves to be the best performing algorithm for the prediction of customer churn.