Customer Churn Analysis

Abstract

The term "churn" is used to describe the termination of a contract; hence, customer churn happens when current customers decide not to be clients anymore. Customer retention experts have the difficult problem of predicting churn. However, thanks to developments in AI and ML, this task is now more feasible than ever before. Research has shown that machine learning may be used to predict client attrition. In order to forecast customer turnover and determine which customer attributes significantly affect churn, this thesis set out to create and use a machine learning model. The Swedish insurance provider Bliwa, who was interested in learning more about client churn, collaborated with us on this study. Gradient Boosting, Logistic Regression, and Random Forest were the three models that were used and assessed. To improve the models, Bayesian optimization was used. Using CrossValidation to gauge their predictive effectiveness, we found that LightGBM achieved the highest PR-AUC, indicating that it was the most effective method for this particular situation. Afterwards, a SHAP-analysis was conducted to determine which customer attributes influence the likelihood of customer turnover. A number of client characteristics were shown to significantly impact churn as a result of the SHAP-analysis. Armed with this information, we may take proactive steps to decrease the likelihood of customer attrition.