Desempenho de classificadores na identificação de padrões neurais associados à imagética musical

Abstract

Brain-Computer Interfaces (BCIs) translate neural signals into commands for controlling external devices without relying on motor pathways. In biomedical engineering, these interfaces have emerged as a promising communication alternative for individuals with motor and speech impairments. Music imagery, defined as the ability to mentally evoke a melody, activates specific brain areas and generates identifiable electroencephalographic patterns, making it a promising paradigm for BCIs. This study aimed to compare the performance of the classifiers Linear Discriminant Analysis (LDA), k-Nearest Neighbors (kNN), and Support Vector Machine (SVM) in discriminating EEG signals recorded during musical imagination and resting states. Twelve healthy subjects participated in the study. The signals were preprocessed, transformed into the frequency domain, and subjected to spectral feature extraction. The data were then split into training, validation, and test sets, and evaluated using accuracy, Kappa coefficient, and confusion matrix metrics. SVM showed the best average performance (accuracy = 0.97 ± 0.03; Kappa = 0.94 ± 0.07), followed by LDA (0.93 ± 0.04; 0.86 ± 0.08) and kNN (0.82 ± 0.08; 0.64 ± 0.16). Despite individual variability, the results confirm that musical imagery produces distinguishable and classifiable EEG patterns. The proposed approach shows potential for application in BCIs, especially in assistive projects developed in clinical and academic contexts.