Machine Learning Approach for Music Familiarity Classification with Single-Channel EEG

Abstract

Recognition of familiar music on brainwaves through machine learning (ML) can be instrumental in innovative therapeutic devices that improve memory and communication in dementia patients. In this study, a variety of machine learning algorithms were applied, including Random Forest (RF), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Deep Learning (DL), to EEG brainwaves from a mobile headset's Fp2 channel. EEG data from 20 participants assessing familiarity with 20 Christmas carols were used. For ML methods (excluding DL), particular frequency bands were selected (theta, alpha, low beta, and high beta), and six statistical features were used to train classifiers. In contrast, DL employed spectrograms and 2D convolutional neural networks. 67% accuracy was achieved with SVM using only the kurtosis features. Due to the variability of the participants, individualized training and testing produced an average accuracy of 72.4%. In dementia care, these results suggest promising therapeutic avenues.