This is a study of churn analysis design based on the meta-analysis method used in previous related research in game and telecommunication industries. The aim is to facilitate the general use of churn analysis in game development.

This study focuses on the statistical modeling and machine learning techniques used to analyze player churn by researchers in game and telecom studies over the past ten years. This study uses the meta-analysis methodology for the literature review. It will provide recommendations as to what techniques are suitable to perform churn analysis. The final section applies Network Analysis and Belief Bayesian Network to data from the game, Wuzzit Trouble, in order to prove that these methods, adopted from the telecom industry, can be used to perform churn analysis in the game industry.

The paper consists of five significant parts. 1) Reviewing the chosen studies to summarizing the most popular techniques in churn analysis to deal with the specific types of datasets or research problems. 2) Analyzing the similarities and differences between data manipulation techniques in churn analysis used in game industry and the other industries. 3) Picking the proper data manipulation techniques for churn analysis in educational games by meta-analyzing the selected studies. 4) Providing a full research plan including research questions, research approaches and research processes for Churn analysis of game such as the Wuzzit Trouble. 5) An implementation on the sample data from the game Wuzzit Trouble.

The investigation in this paper addresses the churn definition, feature selection, data processing statistical models and the machine learning techniques, especially statistical graphics model, to present the analysis on churn problem as well as the possible relationship between each variable. The contribution of the paper provides an analytical research design of the churn analysis graphically, which has never been utilized in churn behavior analysis in the game industry previously. The implementation of the techniques mentioned in the research design, the result of the paper, proved its applicability by taking the Wuzzit Trouble data as the sample. The methodologies include Discriminant Analysis, Collinearity Test, Network Analysis on features and Bayesian Belief Network visualization on churn probability.