Feature Selection component - ClickML Studio

The Feature Selection component removes irrelevant or redundant features. Fewer, better features typically lead to faster training and more generalizable models.

Configuration

Option	Description	Default
Method	Selection algorithm (see table below)	—
Target Column	The column being predicted (required for supervised methods)	—
K	Number of top features to keep (supervised methods)	`10`
Threshold	Variance or correlation threshold (unsupervised methods)	`0.0` / `0.9`
Estimator (RFE)	Model used internally to rank features: `Random Forest`, `Logistic Regression`, `Linear Regression`	`Random Forest`

Methods

Method	Type	How it ranks features
Variance Threshold	Unsupervised	Drops features whose variance is below the threshold
Correlation Threshold	Unsupervised	Drops one of each pair of features correlated above the threshold
Select K Best (Chi2)	Supervised	Ranks features by chi-squared statistic (non-negative values only)
Select K Best (F-score)	Supervised	Ranks features by ANOVA F-score
Select K Best (Mutual Info)	Supervised	Ranks features by mutual information with the target
RFE	Supervised	Recursively removes the least important features using an estimator
Lasso (L1)	Supervised	Drops features whose Lasso coefficient is zero

Input / Output

	Type
Input	DataFrame
Output	DataFrame (selected features only)