Logistic regression is a statistical model used to predict categorical outcomes. It can be used for classification problems with two or more classes, such as predicting if a customer will return or not return. The model estimates the probabilities of the different classes and uses a cutoff to classify new observations. Some key applications of logistic regression include classifying customers, differentiating factors between groups like male and female executives, and predicting loan approvals. It focuses on binary dependent variables coded as 0 and 1.