Harnessing Residual Attention Networks for Stress Level Classification Using EEG Spectrograms

Stress detection is becoming one of the important applications in neuroscience and mental health. Conventional measurement of stress levels is often done by self-reported methods such as through filling out questionnaires, which rely heavily on individual perceptions and tend to be subjective because they are influenced by personal biases. To overcome this problem, a more objective approach is proposed by analyzing Electroencephalogram (EEG) signals that capture brain activity. Increasing detection accuracy is now supported by many developments in machine learning. This study focuses on stress level classification using spectrogram images from EEG signals. The main objective of this study is to develop a detection model that can identify stress levels based on EEG data. In this study, EEG signals were processed through several steps, including segmentation, filtering, and conversion into spectrograms. Furthermore, the Residual Attention Network model was tested to classify stress levels. The results of this study indicate that the spectrogram has excellent potential in recognizing stress patterns. This study produced a training accuracy of 0.77 with a testing accuracy of 0.53. These findings suggest that integrating more sophisticated architecture and preprocessing methods could significantly enhance the accuracy and reliability of stress detection models. This improvement has the potential to lead to more effective and timely interventions in mental health care in the future.