Prediction Model For Replacement Of Smartphones With Mobile APP Usage

Smart phones have exceeded personal computer,becoming the most important way to access the Internet,with the number of APP applications showing spurt growth.The type,frequency,duration and flow of APP application are closely related to the basic attributes and behavioral preferences of users.Therefore,mining mobile phone usage model is not only helpful to optimize the APP application,enhance the mobile phone operating system interface design,but also to stimulate a large number of intelligent business applications,such as target customer segmentation,word-of-mouth marketing,target advertising,and so on.In this paper,based on the mobile phone log data of a certain operator in Jiangsu Province,in order to identify target customers need to replace the intelligent mobile phone,we have conducted in-depth research,including data modeling of APP usage and prediction model construction of replacement smart phones.Replacement predicting model can provide operators with a replacement demand of customer segments,which help to carry out precision marketing,has a huge potential commercial value.In detail,the main work of this paper includes the following aspects:1.Feature analysis of mobile phone log data.In order to effectively classify the user attributes,application types and usage,the Temporal Bag-of-Apps data model is proposed in this paper.We explore user annotation models and specific APP application amount in users Internet log data and define the user's replacement behavior and replacement time,especially,consider the use of time,influence on the replacement behavior,the time is represented as discrete time,using in data modeling.2.A prediction model based on Cox proportional hazards regression is proposed.This is an effective multivariate survival analysis method to analyze the influence of many factors on survival time.This model consists of survival time and risk function and to predict whether a user replacement,the regression coefficients of the covariate have been used in understanding the various factors on the influence of replacement behavior.This model explains in theory to predict the most natural replacement with self-learning function,which can be the decisive factor in adaptive learning and update replacement events.3.In this paper,we predict the key attributes by partial regression coefficient and get the risk function and the survival time of each user.Then compared with other classification models and actual observation data to get the prediction accuracy.The recall of our risk model is much higher than that of other models,which leads to the model with the best overall performance.At the same time,we also found that the use of APP in the evening stage has a weak discriminative power for different users.Users who use mobile phones at night are more likely to change their mobile phones,and the application is very trendy,these users are likely to be young.In addition,many traditional applications have a negative impact on smartphone replacement behavior.This result is of great significance to management,which should be taken into account the characteristics of the user itself and more generally observe the user groups,in order to get more general group characteristics,which needs further study.