The Machine Learning “Advent Calendar” Day 12: Logistic Regression in Excel

Logistic regression is explained as a classification model for binary outcomes, contrasting its statistical and machine learning perspectives. It demonstrates why linear regression is ill-suited for classification tasks and introduces the sigmoid function to constrain outputs between 0 and 1. The training process involves minimizing a log-loss function, derived from Bernoulli likelihood, using gradient descent to find optimal parameters. The entire process, from calculating gradients to updating parameters, is illustrated through a step-by-step implementation in Excel.