Analysis Of IPhone Order Data Based On Neural Networks Model

In recent years, the market competition is becoming increasingly fierce, operators try to expand market share. The operators go out of their ways in the user segment, and to minimize the cost of advertising because of the acceleration of competition. And in this context, marketing plays a role that cannot be ignored. In the case of marketing on a budget, we can collect users'data, build a data model, thus we can locate the target user more accurately, and promote in the most appropriate channels with the most suitable product, and correspondingly again higher possibility of successful marketing.This paper's main body is the mobile users who has participated in the appointment of Apple which was launched by Zhejiang Mobile, and this paper's object of study is to predict the reaction of mobile users to the appointment of Apple, and make those mobile users who appoint Apple successfully digest it, that successfully purchase. The data set, contains nearly 6 million observations came from different databases of Zhejiang mobile platform, with phone numbers be unique recognized identifier matches data to get all the observations. In the data set, two main parts can be distinguished:appointment data and user data, with the former including users'basic information, and the phone's related property information; the latter including users whether eventually purchase the phone they appointed, users' age, number attribution, number brand, and whether APP user, the times user login in On-line Business Hall and so on, total 24 variables, and covers two kinds of variable continuous type and categorical type.In order to mine commercial information from the existing data as much as possible, this paper uses the analysis of variance to analyze the effect of phone's attribute on the number of appointment, and uses Logistic regression model and neural network model to build data model for user data, in the first stage uses Logistic regression model to forecast users'response rate, and filter out variables who have significantly contribution on the response rate through stepwise regression method; in the second stage we use the 10 variables filtered out in the first stage by using Logistic regression model and user whether purchase the phone or not build neural network model to predict the user's response rate. Of course, in order to make the model is robust, we divide the raw data into two data sets, training set and testing set. By observing the ROC curve and the confusion matrix to test if this model are validated. We found models in both training and testing sets the fitting effect are very good, which suggested the two models are robust, and they are effective models to forecast the users'response rate.