Clustering, PCA, and Model Evaluation

Planned scope¶

• feature scaling and geometric intuition;

• PCA as a controlled dimensionality-reduction method;

• k-means and hierarchical clustering;

• confusion matrices, ROC curves, and feature importance where appropriate;

• comparing an independently implemented method with a library implementation.

The goal is not to treat machine learning as a black box. It is to judge what a model has learned, how it can fail, and whether that result supports the scientific claim.