Comparative Data Resample to Predict Subscription Services Attrition Using Tree-based Ensembles

The digital market today, is rippled with a variety of goods/services that promote monetization and asset exchange with clients constantly seeking improved alternatives at lowered cost to meet their value demands. From item upgrades to their replacement, businesses are poised with retention strategies to help curb the challenge of customer attrition. Such strategies include the upgrade of goods and services at lesser cost and targeted improved value chains to meet client needs. These are found to improve client retention and better monetization. The study predicts customer churn via tree-based ensembles with data resampling such as the random-under-sample, synthetic minority oversample (SMOTE), and SMOTE-edited nearest neighbor (SMOTEEN). We chose three (3) tree-based ensembles namely: (a) decision tree, (b) random forest, and (c) extreme gradient boosting – to ensure we have single and ensemble classifier(s) to assess how well bagging and boosting modes perform on consumer churn prediction. With chi-square feature selection mode, the Decision tree model yields an accuracy of 0.9973, F1 of 0.9898, a precision of 0.9457, and a recall of 0.9698 respectively; while Random Forest yields an accuracy of 0.9973, F1 of 0.9898, precision 0.9457, and recall 0.9698 respectively. The XGBoost outperformed both Decision tree and Random Forest classifiers with an accuracy of 0.9984, F1 of 0.9945, Precision of 0.9616, and recall of 0.9890 respectively – which is attributed to its use of hyper-parameter tuning on its trees. We also note that SMOTEEN data balancing outperforms other data augment schemes with retention of a 30-day moratorium period for our adoption of the recency-frequency-monetization to improve monetization and keep business managers ahead of the consumer attrition curve.