Convolutional Neural Network for Octave Illusion Classification

Abstract

The octave illusion occurs when two tones with one-octave differences are alternately played to both ears repeatedly. This study aims to classify participants into illusion and non-illusion groups by applying a convolutional neural network. Brain activity data were recorded using magnetoencephalography (MEG), and the activation levels between the two groups were analyzed. This study proposes a method for developing several layers of learning units to compare activities in the same brain region for the illusion and non-illusion groups. This study is one of the first attempts to apply deep neural networks for the classification of MEG data to illusion and non-illusion groups. The developed convolutional neural network showed stable results in the classification of octave illusion and non-illusion data with 100% accuracy and low training and validation losses, which indicate that no overfitting occurred. Furthermore, the pre-trained, octave illusion dataset convolutional neural network showed promising results in a similar auditory illusion data classification and can be used as a universal tool for classifying auditory illusions using MEG data.