Summary
We started with a simple linear regression model for the binary classification problem and saw t how limited it was. The probit regression model, which is an adaption of the linear regression model through a latent variable, overcomes the drawbacks of the straightforward linear regression model.
The versatile logistic regression model has been considered in detail and we considered the various kinds of residuals that help in the model validation. The influential and leverage point detection has been discussed too, which helps us build a better model by removing the outliers. A metric in the form of ROC helps us in understanding the performance of a classifier. Finally, we concluded the chapter with an application to the important problem of identifying good customers from the bad ones.
Despite the advantages of linearity, we still have many drawbacks with either the linear regression model or the logistic regression model. The next chapter begins with the family of polynomial regression...
