Identifying the customer churn drivers in the gaming ecosystem using machine learning models

Abstract

In the highly competitive game industry, retaining players and minimizing churn is essential for a game company’s long-term success. However, identifying the factors that drive customer churn is challenging, especially with dynamic player behaviors, particularly as the triggers of spenders and non-spenders. This study aims to explore the key drivers of customer churn and develop effective retention strategies using machine learning techniques on sequential player data. A full pipeline was implemented, including three stages: (1) a sequence generator to transform raw sequential data into a format suitable for recurrent neural network (RNN)-based input; (2) an RNN-based model to learn temporal patterns in user behavior; and (3) a downstream deep reinforcement learning (DRL) model to identify the key drivers of churn and recommend optimal retention strategies. Training was conducted separately for spenders and non-spenders to account for their behavioral differences and to prevent data bias. Sequence statistics were used to determine appropriate sequence lengths for each group, which guided model training. The gated recurrent unit (GRU) model for feature extraction achieved an accuracy of 0.94 for non-spenders and 0.78 for spenders. Finally, the DRL model provided valuable insights into the behavioral factors influencing churn, upon which actionable retention strategies were developed.