Human emotional stress analysis through time domain electromyogram features

Emotional stress is induced on any person who is subjected to experience negative emotions (disgust, anger, fear and sadness) over a prolonged duration. Physiological signals (electrocardiogram (ECG), electromyogram (EMG), galvanic skin resistance (GSR), etc) effectively reflect the behavior of emotional stress compared to other modalities. In this work, the proposed experimental protocol induces emotional stress through audio-visual stimuli (video clips) and simultaneously acquired the EMG signals. EMG signals are preprocessed using IIR Notch filter, Elliptic filter and Discrete Wavelet Transform (DWT). Time domain features such as simple square integral (SSI), integrated EMG (IEMG), waveform length (WL) and difference of absolute standard deviation value (DASDV) are extracted from EMG signal and one-way ANOVA is used to compute the significance of each feature. In the classification stage, the subjective emotions are classifiers into negative (sadness, disgust, fear and anger) and positive (happiness and surprise) emotions. Thereafter, the negative emotions are further classified into emotional stress and non emotional stress. Two non-linear classifiers such as K Nearest Neighbor (KNN) fuzzy K Nearest Neighbor (fKNN) are used on this work. WL is founded as the most prominent feature among all four extracted features on emotional stress classification. The maximum mean classification rate of 70.85 % is achieved using KNN. According to the sensitivity values, KNN is better in identifying negative emotions and also the subject emotional stress.