fMRG Verilerinde Temel Bileşenler Analizi ve Özyinemeli Boyut Eliminasyonu Kullanarak Boyut Küçültme

In this study, dimension reduction analysis is done on the Functional Magnetic Resonance Imagining (fMRI) data. The reduction of voxels which are the dimension in our case is the fundamental step in developing of a generalized model. To reach this goal, two different methods have been applied. In the first one Principle Component Analysis (PCA) is used to reduce the effect of curse of dimensionality. On the other hand, the method known as Recursive Feature Elimination (RFE) is used to drop the voxels with less discriminative information. RFE ranks the voxels according to their weights in the model obtained from Support Vector Machine, then eliminate the voxels with low rank. The obtained result showed the outperforming of PCA over RFE. But, due to the transformation of new space, the obtained dimensions at the output of PCA do not contain the 3D coordinate information. Therefore, RFE can useful when the selected voxels are interested such as neuroscientifical and psychology studies.