LDA

LDA

Linear Discriminant Analysis - supervised dimensionality reduction that maximizes class separability.

When to use:

• Have labeled data (supervised)
• Want to maximize class separation
• Feature extraction before classification
• Need interpretable discriminant functions
• Visualization of class structure

Strengths: Supervised (uses labels), maximizes class separation, fast, interpretable Weaknesses: Linear only, requires labels, max components = n_classes - 1

Model Parameters

Feature Columns (required) Input features for dimensionality reduction.

Target Column (required) Class labels for supervised learning.

N Components (optional) Number of discriminant components.

• null: Use min(n_features, n_classes - 1) (default)
• Custom: Between 1 and n_classes - 1

Solver (default: "svd") Method to compute components:

• svd: Singular Value Decomposition (default, recommended)
• lsqr: Least squares solution (for many features)
• eigen: Eigenvalue decomposition (classic method)

Shrinkage (optional) Regularization for covariance estimation:

• null: No shrinkage (default)
• auto: Automatic shrinkage (good for small samples)