Removing redundant features via clustering : preliminary results in mental task separation

Recent clustering algorithms have been designed to take into account the degree of relevance of each feature, by automatically calculating their weights. However, as the tendency is to evaluate each feature at a time, these algorithms may have difficulties dealing with features containing similar information. Should this information be relevant, these algorithms would set high weights to all such features instead of removing some due to their redundant nature. In this paper we introduce an unsupervised feature selection method that targets redundant features. Our method clusters similar features together and selects a subset of representative features for each cluster. This selection is based on the maximum information compression index between each feature and its respective cluster centroid. We empirically validate out method by comparing with it with a popular unsupervised feature selection on three EEG data sets. We find that ours selects features that produce better cluster recovery, without the need for an extra user-defined parameter