Logistic Regression

Logistic regression predicts a yes/no outcome — will this customer churn, is this email spam — by outputting a probability between 0 and 1 rather than an unbounded number. It does this by passing a linear combination of the inputs through an S-shaped curve (the logistic function) that squashes any value into the 0-to-1 range.

Despite the name it is a classification tool, not a method for predicting quantities. You turn its probability into a decision with a threshold (often 0.5, but you should tune it for the costs at hand), and you interpret its coefficients on the log-odds or odds-ratio scale rather than as direct changes in probability.