Logistic regression
Also: logit model, softmax regression
Despite its name, a linear classification model: p = σ(w·x + b), trained by minimizing binary cross-entropy. Its decision boundary is the hyperplane where p = 0.5; the multiclass version uses softmax.
p(y = 1 | x) = σ(w·x + b) = 1 / (1 + e^−(w·x + b))
Example
If w·x + b = 2 for an email, the model gives σ(2) ≈ 0.88 probability of spam, so it is classified as spam at the usual 0.5 threshold.