Evaluating Deep Learning with different feature scaling techniques for EEG-based Music Entrainment Brain Computer Interface

Abstract

Music Entrainment Brain-Computer Interface (BCI) systems influence music as a modulatory tool, synchronizing neural activities to the rhythm and structure of auditory stimuli. This innovative interface uses electroencephalogram (EEG) signals to decode cognitive states influenced by music, enabling novel pathways for cognitive enhancement, stress reduction, and therapeutic interventions. This paper investigates the influence of different feature scaling techniques on the performance of deep learning model within EEG-based Music Entrainment Brain-Computer Interface (BCI) systems. Deep neural network (DNN) is implemented to classify EEG signals into three classes i) during listening to music, ii) during listening to singing bowl sounds and iii) relax states. The comparison on effect of both music and singing bowl therapy on different brain lobes, such as the frontal, temporal, central and occipital lobes are analyzed. The DNN model performance is evaluated by employing various feature scaling methods like StandardScaler(), MinMaxScaler(), Normalizer(), and RobustScaler(). DNN model loss is computed using four loss functions such as mean squared error(MSE), mean absolute error (MAE), logcosh and categorical cross entropy. StandardScaler with mean absolute error loss function showed test accuracy of 87.26%. This research offers valuable insights into BCI's potential in stress management and the integration of music as a therapeutic tool.