If you have labelled data, you can try linear discriminant analysis (LDA; also known as canonical discriminant analysis), which maximizes the between-class vari

If you have labelled data, you can try linear discriminant analysis (LDA; also known as canonical discriminant analysis), which maximizes the between-class variance while minimizing the within-class variance to best separate different classes by projecting data onto a new space that maximizes class separability (for whichever classes help your specific use case).