Dimensionality reduction

• high dimensional space analysis is hard

Feature selector

• Select few dimensions
• Feature == dimension
• Create new feature from several other features

Prinicpal component analysis (PCA)

• Remove linearly dependent dimensions from data
• Gives intrinsic dimensionality
• Linear combination of multiple dimensions
• Maximize variance
• minimize the sum of squared errors (sse)
• mean point is a point that minimizes the sum of squared errors
• Line that maximizes variance and minimizs the sum of squard errors