1. Classification problems
ex) E-mail: Spam or Not Spam
ex) Tumor: Malignant(positive class, 1) or Benign(negative class, 0)
1.1 Linear regression algorithm
First, Let's apply Linear regression algorithm.
Ex 1) In the below case, The Linear regression seems to work quite reasonably.
Ex 2) But through below case, You could find out that Linear regression is not an appropriate algorithm to resolve Classification problems.
Also, In the Classification problems, y is either zero or one. But If we apply Linear regression, the hypothesis could output can output values which are much smaller than zero or bigger than one. (Even if all of your training examples have labels y equals zero or one.)
∴ That is why we need Logistic regression.
1.2 Logistic regression
∴ From now on, we will use Logistic regression as a Classification algorithm.
* Do not confuse with Linear regression because of the common word (regression)!
You could see that predictions of logistic regression are always between zero and one. It cannot become bigger than one or smaller than zero.
Comentários