I recently got asked with this question and realized that there are some misconceptions about linearity.
Logistic Regression uses the sigmoid function to convert the output into probabilities between 0 and 1.
This sigmoid function is very much non-linear.
So does that make the Logistic Regression model non linear?
NO.
Linearity occurs when the interaction between weights(W) and predictors(X) of the regression equation is additive:
W1X1 + W2X2
There are no interaction terms of the weight predictors here like W1X1 * W2X2 which would make the model non-linear.
So it doesn’t depend if the sigmoid is non-linear. The model doesn’t depend on it.
The model depends on the regression equation beneath it which is very much linear.
This equation is what that forms the decision boundary or the hyperplane which will still be linear.
What’s your take on this?
#machinelearning #datascience