Redundant features are those that are relevant to the target concept, but their removal has no negative effect. Usually, a feature becomes redundant when it can be expressed by other features. Redundant features unnecessarily increase data dimensionality [89], which worsens the learning performance. It has been empirically shown that removing redundant features can result in significant performance improvement [69, 40, 56, 210, 6, 43]. In the last section, we introduced the SPEC framework for spectral feature selection. We notice that the feature evaluation criteria in SPEC are univariate: features are evaluated individually, therefore the framework is not capable of handling redundant features.