K-Nearest Neighbors Classifier component - ClickML Studio

K-Nearest Neighbors (KNN) makes predictions by finding the K training examples closest to the input and returning their most common class. No training phase is required — the model stores the training data and queries it at prediction time.

Configuration

Parameter	Description	Default
N Neighbors	Number of neighbors to consider.	`5`
Weights	`uniform` — all neighbors vote equally. `distance` — closer neighbors vote more.	`uniform`
Metric	Distance metric: `minkowski`, `euclidean`, `manhattan`, `chebyshev`	`minkowski`
P	Power parameter for `minkowski`. `p=1` is Manhattan, `p=2` is Euclidean.	`2`

Input / Output

	Type
Input	`X Train` + `Y Train`
Output	Trained Model

When to use

Simple and effective for small datasets. Always scale features — KNN is purely distance-based and will give misleading results if features are on different scales. Slow at prediction time on large datasets.