k plus proches voisins

k-Nearest Neighbors classifies a point by looking at the examples closest to it. Birds of a feather end up in the same class. 🐦

kNN is a lazy learner: it just memorizes the data. The real work happens later, when a new point asks for a label.

To find who is near, the model computes a distance to every training point, most often plain straight-line distance.

The closest k neighbors each cast a vote. Whichever class wins the majority becomes the prediction for the new point.

The letter k is how many neighbors you consult. It is the one knob that most shapes how kNN behaves.

A tiny k, like 1, follows every wiggle in the data and reacts to noise, giving jagged, overfit boundaries.

A big k averages over many points for smoother decisions, but too large and it blurs the real differences between classes.

Distance is unfair if one feature spans thousands and another spans tenths. Always scale features so each counts equally.

Create it like any estimator and pick your n_neighbors right away. Here we consult the five closest points.

The familiar contract still applies: fit stores the data, predict finds neighbors and votes on the answer.

kNN is simple but can be slow at prediction time, since every guess compares the new point to all stored examples.

Let's pin down the parameter that defines kNN.

kNN predicts by letting the closest k neighbors vote. Choose k with care and scale your features, and it works well. 🎯