engleski

Churn in the mobile gaming field: Establishing churn definitions and measuring classification similarities

Churn prediction gained attention across different application fields, both in the business and academic world, and a variety of sophisticated churn prediction models have already been built. One of the most important issues with churn is the lack of a good definition. This is especially prominent in non-contractual business settings where a definition of a churner needs to be stated prior to building the churn prediction model. The churn definition statement is highly subjective and will necessarily influence the model and the resulting classification. This study raises the problem of the churn definition statement and proposes four groups of churn definitions relying on user activity or user engagement. Although proposed definitions are established within the field of freemium mobile games, they can be applied to various non-contractual business settings, as they are related exclusively to user behavior. When applied to real data, different definitions will yield inconsistent classifications. Consequently, the second problem addressed in this research is evaluating similarities between churn definitions. The methodology for evaluating the similarity of churn definitions and churn definition groups is proposed following the principle that definitions applied on a set of users are more similar if they induce more similar churn classifications. Similarities in churn definitions and corresponding classifications are evaluated by applying the Jaccard similarity coefficient and its k-adic formulation. Comparing similarities between definition groups is a more challenging task since within-group heterogeneity must be taken into account. To deal with this problem this paper proposes a modified 2-group k-adic Jaccard similarity coefficient that can be applied in different fields, beyond just churn classification. Proposed churn definitions and similarity measures are applied on a real dataset as a case study that further illustrates the developed methodology and its applications in real-life problems.

Churn definition ; Similarity coefficient ; Resemblance measures ; Game data mining

nije evidentirano