Research About Online Shopping Users’ Churn Based On Data Mining

With the development of the Internet, E-commerce has more obvious advantages than traditional shopping pattern. Shopping online this convenient and fast shopping model is attracting more and more people. At the same time, such a large scale of transactions and demand accelerate the competition and elimination among online retailers. In promoting the development of E-commerce, but also to accelerate the survival of the fittest of online retailers. The competition among E-commerce enterprises has been intensified.Not only to make their products attractive, The customer has become the most important resource for enterprises. How to keep high quality customer has become the focus of the work of the enterprises.The customer churn problem has caused more and more enterprises’ attention. In order to ensure the healthy development of their own in the fierce market competition, not only to make their own products attractive, but also deeply understand the user’s preferences and satisfaction, Deeply extracting user’s behavior characteristics. E-commerce users’ behavior has instability and high churn rate, How to find them before the churn of customers and help the marketing department determine the target customer group, making effective policy to retain valuable customers. So, whether online retailers can correctly predict the churn rate or not decides retailers’ success or failure in the market. In these areas, data mining can help companies.In this paper, data mining is applied to the business analysis, trying to predict the buying behavior of the lynx users, to determine the loss of users in a specific period, making some retention measures to reduce the loss rate.This paper mainly carried out the following three aspects of the work.Firstly, This paper introduces the research background and the current theoretical research methods of customer churn and three models using data mining to predict the customer churn, Including: decision tree model, gradient boosting decision tree model, referred to as GBDT, logistic regression model.Secondly, this paper use GBDT model to construct the feature set and build logistic model to predict and label the sample.Thirdly, this paper get Tmall web users’ four months’ data as a real data set. The purpose is to predict the purchase behavior of Tmall users for the fifth month,and then, making an analysis of the empirical results of the model.This paper obtained the good prediction results in the empirical analysis. The accuracy obtained is 86%, the recall rate is 79%. The model has a good predictive ability that can be applied in the E-commerce enterprises’ daily customer churn prediction. It can really help the electronic commerce enterprise to strengthen customer management and improve customer retention. The research method used in the analysis is expected to be adopted in other enterprises.