Recall rate: Suppose we give a business department a category email list list of 20 lost users through data mining. As a result, 16 people in the list have indeed lost. The precision rate of this model reaches 80%, which is quite good, but the problem is that users are lost in the end. It was 1,000. At this time, the business department quit, and the results predicted by the model were thousands of miles away from the actual business scenario. At this time, an indicator that the model needs to introduce is the recall rate, which is also called the model coverage rate, that is, after the model inputs big data, it can category email list more comprehensively cover the lost users we need to find.
During the early warning period, we finally established a churn category email list prediction model through sample data, and continuously trained the data to effectively improve the precision and recall of the model. Next, we can predict the list of user churn in the next month or next quarter. We need to import the full user data into the model, obtain a scoring set or rule set of churn rules, and mark users for accurate hierarchical operations. 2. Building a loss early warning model In order to better illustrate the idea of model building, we create a virtual case: a telecom operator needs to category email list predict the probability of user churn through the user churn early warning model in order to reduce customer churn.
We use virtual data to determine the modeling sample data, and category email list filter the following data Field: What is the essence of lost user operations? Analyze how to build a user loss early warning system from three aspects We start training the model after getting the sample data. There are three algorithms in terms of modeling ideas for user churn warning: cox survival model: the biggest role of this model algorithm is to analyze the relationship between each user category email list variable and churn, and predict the probability of different users churn in the future through the survival algorithm. Decision tree mode.