Examining Brainwave Patterns in Response to Familiar Music: An EEG and Machine Learning Approach

Detecting familiar music using brainwaves through machine learning can be pivotal for innovative therapeutic devices benefiting dementia patients' memory and communication. We applied various machine learning algorithms, including Random Forest (RF), Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), K-Nearest Neighbor (KNN), and Deep Learning (DL), to EEG data from a mobile headset's Fp2 channel. EEG data from 20 participants assessing familiarity with 20 Christmas carols was collected. For ML methods (excluding DL), specific frequency bands were extracted (theta, alpha, low beta, and high beta), and six statistical features served as input for training ML classifiers. DL method used spectrograms and 2D convolutional neural networks. SVM with only kurtosis features reached the highest accuracy at 66.9%. Since there are variations among the participants, individualized training and testing resulted in an average accuracy of 72.3%. These results suggest promising avenues for therapeutic strategies in dementia care.