EEG Signal Classification for Mental Stress During Arithmetic Task Using Wavelet Transformation and Statistical Features

This paper on electroencephalogram (EEG) analysis describes a feature extraction model based on discrete wavelet transforms for the classification of EEG signals during a mental arithmetic task as mental stress. Stress is a normal part of life. Stress analysis is more important than sleep analysis because stress causes sleeping disorders. The EEG is used to monitor the electrical activities of the brain. The features were computed in time-frequency domain using wavelet decomposition and they were extracted from single-channel EEG signal during and before a mental arithmetic task. The amplitude, power, and frequency of extracted decomposed signals changed significantly during mental stress compared to the state before the mental arithmetic task. 12 state-of-art statistical features were computed from each decomposed signal (5 decomposed frequency bands gave a total of 60 features) of a fixed window of 10 seconds. The Neighborhood Component Analysis (NCA) feature selection method was used to find the most significant features, which resulted in the discovery of four features. Finally, mental stress was classified using the selected four features and Ensemble Subspace KNN classifier. The classifier's performance was evaluated on 11 subjects using a five-fold cross-validation technique, which crossed state-of-the-art methodologies in the literature by 77.3 percent.