7. Understanding Classification Models and Assessing their Performance

A classification model aims to place a record into the appropriate category, or group.  In marketing analytics, common classification questions include:  Will this customer renew her membership?  Will that customer redeem the discount coupon?  Will a family of five, with three young children, living in the Connecticut suburbs of New York City, react favorably to this marketing campaign? 

The classification problems described above are binary, as they have two possible outcomes.  

Keep in mind that classification problems are not always binary.  An optical character recognition model that reads numeric characters would have 10 different outcome classes (0 through 9).  A model that predicts a customer’s level of satisfaction, rated as “Poor”, “Fair”, “Good”, or “Excellent” would have four unique outcome classes. 

Note also that a classification model with several outcomes can be converted into a binary scenario.  To make this happen, the modeler needs to identify one class of interest, and then combine the other category outcomes into a general “other” category.