Extended RFM logit model for churn prediction in the mobile gaming market

Abstract

As markets are becoming increasingly saturated, many businesses are shifting
their focus to customer retention. In their need to understand and predict
future customer behavior, businesses across sectors are adopting data-driven
business intelligence to deal with churn prediction. A good example of this
approach to retention management is the mobile game industry. This business
sector usually relies on a considerable amount of behavioral telemetry data
that allows them to understand how users interact with games. This
high-resolution information enables game companies to develop and adopt
accurate models for detecting customers with a high attrition propensity.
This paper focuses on building a churn prediction model for the mobile
gaming market by utilizing logistic regression analysis in the extended
recency, frequency and monetary (RFM) framework. The model relies on a large
set of raw telemetry data that was transformed into interpretable
game-independent features. Robust statistical measures and dominance
analysis were applied in order to assess feature importance. Established
features are used to develop a logistic model for churn prediction and to
classify potential churners in a population of users, regardless of their
lifetime.