1

As I am studying Logistic regression, I just wanted to confirm that I understand it correctly, my main concern is the equation wx+b.

here is my understanding :

(logistic regression is an algorithm used to classify things. The goal is to classify. So let’s assume that we are classifying if something belongs to class 0 or class 1 . We use a function that gives us a probability of it belonging to a class 0 or 1. We set a threshold such as 0.5, if the prob above 0.5 it belongs to class 1 and vice versa. The function that gives us values between 0 and 1 is called sigmoid. the logistic regression uses decision boundary to decide to which class the data belong to, this decision boundary has the weights w and has the formula of : wx + b or w1x1+w2x2+ b )

or are the weights for the sigmoid and not for the decision boundary ?

Mohammed Younis
  • 11
  • 1
    think of it like this. wx + b is linear regression. you fit a curve to your data, it will give you a continuous value. when you want to do binary classification, you want a probability distribution. so sigmoid(wx + b) gives you the probability of the positive class. it squeezes the continuous linear output to [0,1]. so the decision boundary would typically be a linear line dividing the sigmoid(wx+b) curve in half. the relationship of w and the logistic curve is not linear. –  Jun 29 '22 at 03:35
  • 2
    Several points in your understanding are not quite correct, starting with "logistic regression is an algorithm used to classify things". It might be interesting to read about Why isn't Logistic Regression called Logistic Classification?. – dipetkov Jun 29 '22 at 12:43

0 Answers0