Emotion Classification in Parkinson's Disease EEG using RQA and ELM

Most of the earlier works focused on diagnosing the Parkinson's Disease (PD) through behavioral measures. Very few researches attempted to identify the emotion impairment in PD through EEG signals. The main objective of this work is to classify the emotions perceived by the PD subjects through audio-visual stimuli using Electroencephalogram (EEG) signals. EEG database of 20 subjects on PD and 20 NC is developed using a 14 channel wireless EEG device at a sampling frequency of 128 Hz. Audiovisual stimuli of six emotions (happiness, sadness, anger, fear, surprise, and disgust) are used to induce the emotions. The acquired EEG signals are pre-processed using the IIR Butterworth filter to remove the noises, artifacts, and interferences in EEG signals and used to derive three frequency bands (alpha, beta and gamma) of EEG data. Recurrence Quantification Analysis (RQA) is used to extract the two most significant features (Maximum Line Length, Maximum Vertical Line Length) from Recurrence Plot (RP). Besides, these features are combined together called ALL features. Therefore, three types of features were tested using one-way analysis of variance (ANOV A) to test its significance in classifying emotions in PD and NC and a five-fold cross-validation method is used to split the features into training and testing set. Finally, the Extreme Learning Machine (ELM) classifier with two different types of kernel functions used to classify the emotions of PD and NC. The maximum mean accuracy of 89.17%, 84.50% is achieved on NC and PD, respectively. The maximum individual class accuracy of NC/PD is sadness - 90.90/87.50, happiness - 91.10/84.30, fear- 88/84.10, disgust - 88.5/82.70, surprise - 87.4/84.60, and anger - 89.10/83.80. Experimental results indicate that RQA features are highly useful in detecting the emotions in PD compared to other methods and ELM gives the highest mean accuracy compared to other works in the literature.