Régression logistique pour oui/non

Classification predicts a category, not a number. Will this email be spam or not? That yes-or-no answer is what we are after. ✉️

Logistic regression is the classic starting point for two-class problems. It is simple, fast, and surprisingly hard to beat.

Despite its name, logistic regression returns a probability between 0 and 1, like a 0.82 chance this customer will churn.

A curve called the sigmoid squeezes any value into the 0-to-1 range. That is how raw scores become clean probabilities.

You build it just like any estimator: import the class, then create the model object ready to learn.

Show it your features and labels with fit, and it learns the weights that separate the two classes.

Call predict to get the chosen class for each row, returning a clean 0 or 1 for every example.

Want the raw confidence instead? predict_proba gives the probability of each class, so you see how sure the model is.

To turn a probability into a label, you compare it to a threshold, usually 0.5. Above it means yes, below it means no.

You can raise or lower that threshold on purpose. A stricter cutoff catches fewer cases but trusts each one more.

Each feature gets a coefficient. A positive one pushes the prediction toward yes, a negative one toward no.

Let's confirm what logistic regression actually hands back.

Logistic regression turns features into a probability, then a threshold decides yes or no. Simple, fast, and a great first classifier. 🎯