Statistic Case: the Churn Model

STATISTIC CASE N°2 MEMORANDUM (The Churn Model)

Abstract

The Churn Modelling is an example of classification. For instance, a mobile telephone company wishes to predict churning, that is, customers leaving the firm, from demographics ( gender, age, etc.) and consumption habits. It collects data on the past behavior of customers, coding the churners as one and the non-churners as zero. The dependent variable (Y) is the dummy coding churning, and the independent variables (X) are the demographics and the consumption habits. Another well-known application of classification methods, illustrated by the example that follows, is predicting the response to a customer discount policy.

Classification methods are getting increasingly popular, because of the availability of big databases. They are usually presented in a more comprehensive framework, that of data mining, which is the process of exploring databases in search of consistent patterns.

There is a wide range of classification methods. Many of them are implemented in data mining suites, such as Weka or XLMiner. Besides the regression approach of this note, it is worth mentioning the classification trees and the neural networks. Both churn modelling and the example that comes with this note are applications of data mining to Customer Relationship Management (CRM). Other applications in that context are market basket analysis, web clickstream analysis, credit scoring, forecasting TV audience, etc.

Method Section

1) Classification with a regression equation. In binary classification, there are only two classes, which can be coded with a dummy variable. We can use a linear regression equation in binary classification. But, since the predicted values will different from zero and one, we must turn them into zeros and ones. We do this with a cutoff value, assigning to class one those cases for which the predicted value exceeds the cutoff, and to class zero the rest of the cases:

a. ² PRED.VALUE>...