StressDetect: A Deep Learning Approach for Mental Stress Detection Using Time-Frequency Representation of EEG Signals

Mental stress is a prominent issue affecting individuals worldwide, necessitating efficient detection methods for timely intervention and support. Electroencephalography (EEG)based stress detection has emerged as a promising approach due to its non-invasive nature and high temporal resolution. In this paper, we present "StressDetect," a novel deep learning model designed for mental stress detection using EEG signals. Our approach involves generating time-frequency representations of EEG signals through the utilization of complex Morlet wavelets and Fourier coefficients. Subsequently, a Convolutional Neural Network (CNN) architecture is designed and trained on the obtained time-frequency plots, using a series of convolution and pooling layers for feature extraction. A softmax layer is employed for binary classification, distinguishing between stress and non-stress states. The proposed TF-CNN model is evaluated through a series of experiments conducted on subsets of mental tasks, yielding promising results with a maximum accuracy of 97.61% in stress detection. The experimental results showcase the robustness, generalizability and efficiency of the model by outperforming the existing techniques.