A covariance matrix is a square table describing how every pair of features changes together. Positive covariance means two features tend to increase together; negative covariance means one tends to increase as the other decreases; near-zero covariance suggests little shared movement. section 6.2.3
In principal component analysis (PCA), features are first standardized so different measurement scales contribute equally. PCA then computes the covariance matrix and uses its eigenvectors as directions of variation; eigenvalues rank those directions by explained variance. section 6.2.3
Using covariance for dimensionality reduction
For molecular features, PCA can transform the original dataset into coordinates along selected principal components. Retaining only the leading components compresses the data while preserving the dimensions that explain the most variance; choosing too few can discard informative structure. section 6.2.3





