Prediction Of Weibo Member Loss

With the high-speed development of Internet technology,the Internet companies have accumulated TB or even PB level of user data.As the Internet company's great wealth,these data can give the company effective cost savings and increase income if they are used properly.So it is very important for us to dig out the information hidden in the data.Data mining is a technology method through the mathematical model of the data fitting by using computer technology to achieve the model,and finally feedback to the business implementation.In the area of customer churn prediction,there are many useful data mining classification algorithms.In this paper,I use data mining technology in the classification prediction algorithm to predict the membership loss of the social platform SINA microblogging.I will use the data of 26171 microblogging users who will be due to membership date from April 2 to 8,2015.First of all,I use k-means cluster method on users' attributes and active behavior data to achieve user segmentation and identify users who have higher value.Because of the unbalanced sample set,I choose the method of resampling on the category with less sample size.I use 70% of the samples as the training set,and then use the logic regression,decision tree C5.0 and neural network classification algorithms to fit the model,and then use these models to predict whether the members will churn.After comparing these models on some evaluating indicators,it turned out that logistic regression behaves better than the decision tree C5.0 in the ROC curve and Lift.After that,I use cost-sensitive learning to improve the decision tree C5.0 model and then predict the list of churn users.