Hands-On Artificial Intelligence for Cybersecurity

Logistic regression

We have seen that one of the limits of linear regression is that it cannot be used to solve classification problems:

In fact, in case we wanted to use linear regression to classify the samples within two classes (as is the case in spam detection) whose labels are represented by numerical values ​​(for example, -1 for spam, and +1 for ham), the linear regression model will try to identify the result that is closest to the target value (that is, linear regression has the purpose of minimizing forecasting errors). The negative side effect of this behavior is that it leads to greater classification errors. With respect to the Perceptron, linear regression does not give us good results in terms of classification accuracy, precisely ...