Abstract

This paper gives both general linear model (GLM) and support vector machine (SVM) analyses of an experiment concerned with tonality in music. The two forms of analysis are both contrasted and used to complement each other, and a new technique employing the GLM as a pre-processing step for the SVM is presented. The SVM is given the task of classifying the stimulus conditions (tonal or atonal) on the basis of the blood oxygen level-dependent signal of novel data, and the prediction performance is evaluated. In
addition, a more detailed assessment of the SVM performance is given in a comparison of the similarity in the identification of voxels relevant to the classification of the SVM and a GLM.A high level of similarity between SVM weight and GLM t-maps demonstrate that the SVM is successfully identifying relevant voxels, and it is this that allows it to perform well in the classification task in spite of very noisy data and stimuli that involve higher-order cognitive functions and considerably inter-subject variation in neural response.