0

I am very confused about the question: As Sigmoid function is non-linear, then how Logistic Regression works as a Linear Classifier? Can anybody please help?

Mahmudul Hasan Suzan
  • 11
  • 5
    Does it help you to write the logistic regression as $\log\left(\dfrac {\hat p} {1 - \hat p} \right) = X\hat\beta ?$ – Dave Apr 19 '22 at 16:29
  • 2
    See the linear-model tag info: a linear model is linear in the coefficients of the predictors. Even the predictors can be involved in nonlinear transformations. There can be a non-linear mapping from the resulting linear predictor to the outcome. The linear predictor function and generalized linear model entries in Wikipedia provide more details. – EdM Apr 19 '22 at 17:00
  • 2
    @Dave's point is that nonlinear functions can have linear contour lines (level sets). In general, when a function is of the form $f(\beta)=g(X\beta)$ for any function $g,$ then the level sets $f^{-1}(y)$ are all parallel linear subspaces (or unions of them). – whuber Apr 19 '22 at 17:00

0 Answers0