Reaserch On Forecasting The Mobile APP Daily Active User

As the main carrier of P2P investment platform,mobile APP carries millions of user active information and billions of cash flow every day.There is obvious periodic effect and holiday effect in the fluctuation of daily active user of mobile APP represented by P2P business.Therefore,this paper aims to build a daily active user forecasting model by traditional time series method and data mining method,the result of which could provide theoretical basis for product operation and marketing strategies of internet financial enterprises.This paper firstly analyzes the daily active user composition of mobile APP and finds that daily active user could be divided into three parts,which is new users,short-term retained users and regular users.short-term retained users and new users are separated from daily active users by the cumulative retention theory of market launch,after which we obtain the series of daily active regular user.This paper proposes a combined forecasting method based on a divide-and-conquer strategy because of the periodic effect and holiday effect.A multiple seasonal ARIMA model is put forward to forecast daily active regular user series during the non-holiday time.As for holiday effect,a nonlinear mapping solution based on Rprop neural network model and GridSearch support vector regression model is proposed.Therefore,a daily active regular user forecasting model is established by combining these two models.At last we obtain the daily active user forecasted value by merging new users and short-term retained users.The research in this paper shows that the combined forecasting model can fit the linear and nonlinear fluctuation rules of time series well.With the results of experiment,the multiple seasonal ARIMA model can better restore the non-holiday cycle fluctuation level.And in holiday time,the neural network model has better forecasting effect in short holidays,while the support vector regression model has better fitting results in long holidays.I hope that internet financial enterprises could effectively predict the level of daily active user by the research of this paper.With the results of forecasted value,enterprises can better grasp the active trend of users,distribute products reasonably,and strengthen the operation more stable from the inside.