Cognitive state classification using clustering-classifier hybrid method

Classification is a familiar technique used to classify objects. Clustering techniques are employed to segment data into multiple groups. Objects present in clusters exhibit similar characteristics. Machine learning classifiers applied on Functional Magnetic Resonance Imaging (fMRI) data facilitate to classify cognitive states. Clustering methods, such as K-means, Hierarchical, Spectral, and Consensus Clustering Evidence Accumulation (CCEAC) are used to segment brain regions effectively using fMRI data. Selection of voxels as features is a challenging task for cognitive state classification. In this paper, a new hybrid method for classification has been proposed by combining both clustering and classifier. K-means, Hierarchical, Spectral, and CCEAC are applied on StarPlus fMRI data, to obtain clusters. The dissimilarity metrics normalized Mutual Information (NMI), Variation of Information (VI) and Rand Index (RI) are used in CCEAC for selection of optimal clusters. The main focus of the present work is to choose few number of voxels as features by using clustering methods and then built classifier based on fewer voxels. The proposed method has been evaluated on standard Starplus data and has shown good classification accuracy for few number of voxels.