Deciphering Explicit and Implicit Features for Reliable, Interpretable, and Actionable User Churn Prediction in...

Area of Science:

• Computer Science
• Human-Computer Interaction
• Game Studies

Background:

• The online video game industry faces intense competition, driving the need for player retention strategies.
• Machine learning (ML) models are increasingly used to predict player churn, but their 'black box' nature hinders adoption by domain experts.
• Existing eXplainable Artificial Intelligence (XAI) techniques struggle to bridge the gap between complex models and non-technical gaming professionals.

Purpose of the Study:

• To develop a reliable and interpretable solution for predicting player churn in online games.
• To enhance the understanding of ML models for non-technical domain experts like game designers and product managers.
• To propose XAI methods that provide actionable insights and identify key features for player retention interventions.

Main Methods:

• Restructuring ML model inputs into explicit and implicit features to improve interpretability.
• Establishing connections between explicit and implicit features to aid expert comprehension.
• Conducting two case studies, incorporating expert feedback and a within-subject user study, to validate the approach.

Main Results:

• The proposed method effectively predicts player churn while offering interpretable insights into model decisions.
• Domain experts demonstrated improved understanding of ML model features and their implications for player behavior.
• The approach successfully identified crucial features for targeted player retention interventions.

Conclusions:

• The study presents a novel, interpretable framework for player churn prediction in the gaming industry.
• Connecting explicit and implicit features significantly enhances the usability of ML models for non-technical experts.
• The developed XAI approach provides actionable insights, facilitating data-driven decisions for player retention.