Exploiting limited players’ behavioral data to predict churn in gamification

The number of users attracted and engaged in a system dictates the value of the system itself. In gamification, timely detection of churners can produce more successful applications by informing both designers and algorithms. While churn prediction has been extensively studied in entertainment games, gamified systems often implement simpler mechanics, leading to a limited set of features compared to full-featured games. In this work, we studied whether limited players’ telemetry data describing in-game activity can be used to train a Random Forest model for churn prediction in a gamified application. Specifically, we analyzed different approaches for data preprocessing and sampling. Then, data from an online free-to-play (F2P) game was used as a validation set. Results show how in-game activity can be successfully used to predict churn. Moreover, from the tree's visualization and interpretation, we found how players’ likelihood of abandoning the game is proportional to their time investment, both in the game and gamified system.