Exploiting General-Purpose In-Game Behaviours to Predict Players Churn in Gameful Systems

The value of a game is assessed by measuring the intensity of the level of activity of its players. No matter how thoroughly though the design is, the litmus test is whether players keep using it or not. To reduce the number of abandoning players, it is important to detect in time the subjects at risk. In the literature, many works are targeting this issue. However, the main focus has been on entertainment games, from which articulated indicators of in-game behaviors can be extracted. Those features tend to be context-specific and, even when they are not, they are proper of full-featured games, and thus, impossible to adapt to other systems such as games with a purpose and gamified apps. In this preliminary work, we fed to an Artificial Neural Network general-purpose in-game behaviors, such as participation data, to predict when a player will definitively leave the game. Moreover, we study the appropriate amount of information, in terms of players’ history, that should be considered when predicting players’ churn. Our use case study is an on-the-field long-lasting persuasive gameful system.