Bag of Activities for Customer Churn Prediction in e-Book Subscription Domain

The growth of the e-book subscription industry and the increase of the competition in the market caused that the prediction of customer churn has become a very important and challenging issue. This is due to the fact that many reports convict that it is much more expensive to acquire new customers than to retain current ones. This paper focuses on developing a set of the most relevant characteristics in the subscription industry that significantly affect customer churn, primarily based on users’ interaction with the Legimi service. Two versions of the learning set were prepared, which differed in the level of data aggregation (3 months and 12 months). Machine learning algorithms based on decision trees, neural networks and linear classifiers were used to evaluate the author’s dataset. Experiment findings reveal that XGBoost reached the best F1 and accuracy: 91% and 88% respectively. The results achieved allow us to predict potential user churn with a high degree of certainty.