Feature Extraction component - ClickML Studio

The Feature Extraction component applies dimensionality reduction algorithms to transform a set of input features into a smaller number of extracted components. This is useful for reducing noise, speeding up training, and visualizing high-dimensional data.

Configuration

Option	Description	Default
Method	Reduction algorithm to use (see table below)	—
Columns	Numerical columns to reduce. Supports `All Numerical Features`.	All numerical
N Components	Number of output components to produce	`2`
Random State	Seed for reproducibility	`42`
Kernel (Kernel PCA only)	Kernel type: `rbf`, `poly`, `sigmoid`, `cosine`	`rbf`

Methods

Method	Description
PCA	Principal Component Analysis — finds the directions of maximum variance. The most common choice.
Kernel PCA	Non-linear version of PCA using a kernel function.
Truncated SVD	SVD-based reduction that works on sparse matrices.
ICA	Independent Component Analysis — separates statistically independent signals.
NMF	Non-Negative Matrix Factorization — requires all values to be non-negative.

Input / Output

	Type
Input	DataFrame
Output	DataFrame (original non-feature columns + new component columns)

The output columns are named pca_1, pca_2, ica_1, etc. depending on the method chosen. Non-selected columns are preserved.