Predicting OTT Service Churn Rates Using Machine Learning and Apache Spark

This research explores the prediction of churn rates in Over-the-Top (OTT) services using advanced machine learning techniques and make them capable of working properly on large datasets processed with Apache Spark. Effective model tuning is achieved by scoring models in both the Random Forest and Gradient-Boosted Trees classifiers by grid searching candidate hyperparameters and cross-validation. A hybrid model provided by the study that takes average of predictions generated from two classifiers to boost accuracy. This analysis could find the best threshold that should be used to classify instances of one class, hence maximizing a function measuring predictive performance by including this important decision in it. The hybrid approach shows a higher Area Under the ROC Curve (AUC) score than pure DML, which indicates better performance on customer churn prediction. This can highlight important factors for customer retention strategies, and show a predictive model works well in the OTT domain.