Performance Comparisons of GNB, RBF-SVM and NN for Stress Levels Classification Using Discrete Wavelet Discrete Transform

Mental health disorders have risen rapidly in recent years due to the pandemic Covid-19. Normal daily life was impeded, and livelihood was disrupted causing stress to accumulate. Early recognition of stress has become imperative to avoid long-term exposure leading to mental health disorders. The application of electroencephalography (EEG) facilitates the need for stress signal identification. By observing the brain wave pattern, stress-related features can be shown through a graphical representation of the brain-machine interface (BMI) device. However, the complexity and huge amount of data recorded from the brain’s activity make it harder to determine the specific characteristics of stress signals. To overcome that, this study proposed a discrete-wavelet transform (DWT) analysis to extract the stress-related features, and classification was conducted through artificial intelligence (AI) algorithms. The recorded EEG data were preprocessed with segmentation and normalization before being labeled into respective stress stages. After that, the decomposition was carried out using DWT before being classified using Gaussian Naïve Bayes (GNB), radial basis function support vector machine (RBF-SVM) and neural network (NN) for performance analysis. The results show that the NN model achieved higher classification accuracy (79.62%) compared to others. In addition, the precision, sensitivity and F1-score of NN achieve a value of 80% for all features which is higher than the other classifiers.