Analysis And Forecast Of The Customers Of The Listed Companies

With the process of the internationalization, Global brand apparel industry competition is becoming increasingly fierce, and the initiative of the transaction gradually shifted from the enter-prise to the customer. Therefore, the advantage in the competition can be gained only by taken the customer as the center. Customer relationship management theory let us know that 80% of the profits were generated by 20% of customers, and the cost to develop a new customer is 5 times of the cost of maintaining an old one. So in the case of limited marketing resources, customer seg-mentation must be carried out to ensure that the limited marketing resources were allocated to the most valuable customers, in order to improve customer retention, and increase revenue purposes.The data sets in this paper are as follows:A company’s Transaction Data, E-mail campaign data and web activity data. The temporary loss of customers by the time node of Jan 1st 2015 and Jan 1st 2014 were defined as In-time Data and Out-time Data. Bootstrap method was used to extract 20000 customers from each of the data set, and the data selected from In-time Data is divided into Training (70%)and In-time Validation (30%). The data selected from Out-time Data is defined as Out-time Validation. The rates of the Training, In-time Validation and Out-time Validation are 1.62%,1.7% and 2%.The paper restructure the historical data to get more than 2800 variables by the RFM model concept firstly. Then it can get clean dataset by detection of variable meaning, handing of exception, processing and transformation and comparison of missing value. And through the Significance test, correlation test, principal component analysis, variance inflation factor and other statistical methods and combined with the practical significance of variable selection, I can finally selected 7 variables into the model. The model’s correct rate of prediction up to 73% in modeling data, which is be fit of expectations. And the results of the 7 variables and the model were test by ten-decile analysis and Lift Chart method to found the high explain of variables and good sorting classification effect of the model. Later the model were apply to the sample validation set and outside the sample validation to test the model’s stability, and found that the model is very similar to the three data sets, the stability of the model is very good.The results of the three sets of data indicated that the most likely return to the customer is the first category, and then the more likely regression is the second to fourth categories, the fifth categories to tenth types of customers have the lowest possibility of regression. It shows that the model is not only accurate and stable, but also has a significant commercial significance. It can help enterprises to accurately identify customers and gain more benefits.