CHURN TIME PREDICTION ON CASUAL GAME USER

• Main Author: Nurrahmah, Lathifah
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/43465
• Institution: Institut Teknologi Bandung

Churn time prediction, or the time when user stops using a service, can be used as an information to plan churn prevention. On casual game X, churn time is defined as when user does not play game for 10 days in a row. To predict churn time, survival analysis which is mainly used to predict event time, is used. This experiment used CRISP-DM as frame-work and users’ activity log as data source. The methods used are statistic method Cox, survival ensemble with conditional inference tree, SSVM, and Cox Neural Network. Experiment shows that the C-Index and IBS of survival ensemble with conditional inference tree have the best value out of the other methods, which is C-Index with 0.899 and IBS with 0.052. The mean of churn time prediction of the best model shows that users tend to stops playing game X after 6 days. From this information, game developer can plan how to prevent churn, such as weekly challenge.