Logistic regression and classification metrics

In 1958 the British statistician David Cox published “The Regression Analysis of Binary Sequences”. His problem was an outcome that is either 0 or 1, success or failure, and how it depends on explanatory variables. A straight line cannot do it: it predicts values below 0 and above 1.

Cox modelled the log-odds of the outcome as linear instead. The result, logistic regression, is still the first model a data scientist fits to a yes/no question, and the baseline any fraud, spam or churn classifier has to beat.