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

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Machine learning (ML) offers potential for novel therapeutic devices to enhance memory and communication in dementia patients.
• Recognizing familiar music via brainwaves (EEG) is a key area for developing such assistive technologies.

Purpose of the Study:

• To evaluate the effectiveness of various machine learning algorithms in recognizing familiar music from EEG brainwave data.
• To assess the feasibility of using ML-based brainwave analysis for potential dementia care applications.

Main Methods:

• EEG data were collected from 20 participants listening to 20 Christmas carols using a mobile headset (Fp2 channel).
• Machine learning algorithms including Random Forest, LDA, SVM, KNN, and Deep Learning were applied.
• Feature extraction involved specific frequency bands (theta, alpha, low beta, high beta) and statistical features for traditional ML, while DL utilized spectrograms and 2D CNNs.

Main Results:

• Support Vector Machine (SVM) achieved 67% accuracy using only kurtosis features.
• Individualized training and testing, accounting for participant variability, resulted in an average accuracy of 72.4%.

Conclusions:

• Machine learning algorithms can effectively recognize familiar music from EEG signals.
• The findings suggest promising therapeutic applications for ML-driven brainwave analysis in dementia care, potentially improving patient quality of life.