Research On Churn Prediction Of Online Players Based On COX Proportional Hazard Model

With the rise of the Internet,the gaming industry has made great achievements;the competition has become increasingly fierce among game operators.At the same time,the concept of "customer-centricity" has gradually become the focus of enterprises;the cost of acquiring new customers has also been on the rise.It can't be ignored that customer churn prediction becomes an important part of the enterprise,and the identification of game players value has also caused widespread concern.Therefore,this paper predicts the churners of A company's online game player based on the COX proportional hazard model.The research object of the paper is A company's online game player.According to the field research,it finds that there are three problems in the game churn management system: The player churn reference index is too simple;the way to maintain stable period is too single;there is no evaluation method to keep players.According to the actual situation of enterprises,and combining with domestic and foreign literature,the paper put forward to the following reasonable solutions.At first,the paper selected 13 variables from three aspects: performance,players activity,and time interval characteristics,and analyzed the impact of single variables on player churn.Then,13 variables were further selected by the likelihood ratio forward sieve method,after that,the selected variables take proportional hypothesis test;the variables which passed proportional hypothesis test build churn prediction model based on the COX proportional hazards model.Finally,combining with the coefficient of variation method,the paper evaluated players' value be based on the RFM model;The value is the foundation that whether the players is worthy to retention or not.The results show that there are 10 variables that is closely related to players churn.The 10 variables are respectively level,combat effectiveness,total online duration,login frequency,the total amount of recharge,recharge frequency,total consumption,maximum recharge amount,the last interval of recharge amount,the interval between the first recharge time and the registration time.The survival curve is that the probability of survival is gradually decreasing with the life cycle.The players were been divided into eight categories by RFM model;the first and third categories of players should focus on retention that got the highest value evaluation score.