What is Partial Least Squares (PLS) regression?

PLS finds components that maximize covariance between features and target.

Unlike PCA, which ignores y, PLS constructs latent variables that capture directions most predictive of the response variable, making it more supervised.