Decision Threshold

How is the decision threshold chosen?

Answer

In logistic regression, the predicted target value \(\hat{Y}\) should be either 0 or 1. To make this classification, we use the modeled probability \(P(Y = 1 \mid X)\) and apply the following rule:

\[ \hat{Y} = \begin{cases} 1, & \text{if } P(Y = 1 \mid X) \geq 0.5 \\ 0, & \text{otherwise.} \end{cases} \]

This is equivalent to,

\[ \hat{Y} = \begin{cases} 1, & \text{if } \beta^{T}X \geq 0 \\ 0, & \text{otherwise.} \end{cases} \]
Notes and comments

Comment 1: If we had two predictors (features), the decision boundary would be a straight line in the 2D predictor space that separates the data points belonging to different classes. This line will of course depend on the estimated parameters \(\beta\).

Back to collection