Optimal variational mode decomposition based automatic stress classification system using EEG signals

In the current landscape of mental health diagnostics, it is essential to accurately assess stress levels. The utilization of electroencephalogram (EEG) data has proven effective in stress detection, particularly benefiting individuals with impairments by facilitating better interaction with the real world. However, the complexity of multi-component EEG signals poses challenges, necessitating thorough decomposition. Practical parameter selection often leads to issues like mode mixing and signal distortion. This research addresses EEG data challenges by introducing a comprehensive approach that integrates adaptive optimization, channel, and feature selection methods. The process begins with the Laplacian score approach, optimizing channel selection for subsequent investigations. Optimal Variational Mode Decomposition (OVMD) is then employed for parameter optimization, utilizing various optimizers to minimize mean square error (MSE). Additionally, recursive feature elimination cross-validation (RFECV) serves as a feature selection strategy, capturing crucial signal information while preventing model over-fitting through 5-fold variation. Subsequently, an adaptive boosting (AdaBoost) classifier is utilized to accurately distinguish between stress and relaxation phases. The proposed framework exhibits exceptional performance, achieving an overall mean accuracy of 98.35%. The adaptive nature of each step enhances the resilience and accuracy of the diagnostic framework, thereby advancing stress assessment techniques in neurology and mental health. The versatility of our methodology holds promise for improved neurology diagnosis, personalized mental health therapy, and the development of precise stress evaluation instruments.