Human emotional classification using EEG based on discrete wavelet transformation technique

This paper proposes a method to classify a person’s emotional status while watching video clippings. A total of 15 students whose Electroencephalogram (EEG) signals were acquired during watching different categories of movie clippings. The human emotions such as Funny, Sad, Horror, Relax, and Romantic were considered, and short-duration movie clippings were played for 3 minutes duration for each emotion with an EEG signal acquired by using a wearable 14-channel Emotiv Epoc X device. The saline-based felt sensor type of electrodes was used in this study with a sampling rate of 256 Samples per Second. The raw EEG signal was pre-processed by enabling the built-in digital 5th-order Sinc filter and notch filter with a bandwidth of 0.2-45Hz to remove artifacts, power line noise, and signal drift. EEG sub-band energies such as alpha, theta, and beta features were extracted by decomposing the EEG signal into four levels with Daubechies (db4) as the mother wavelet. For every 5s of raw EEG data, the time-frequency features were extracted using discrete wavelet transformation using MATLAB 2022. The relative sub-band energies and band ratios such as heart rate index (HRI), neural activity, vigilance index, arousal index, performance Enhancement index, cognitive performance attentional resource index (CPARI), CNS arousal, and desynchronization were computed. The above 11 features were fed to the machine learning classifier for classification, and their classification accuracy was compared under different emotional data combinations. The result shows that the ensemble algorithm provides the highest classification accuracy of 62.8% for five emotions: Funny, Sad, Horror, Relax, and Romantic. Similarly, a detailed study was presented on comparing under 4, or 3, or 2 combinations of emotional data for classification with different classifier algorithms: decision tree, ensemble, SVM, neural network, and kernel. The highest classification accuracy of 88.80% was obtained for the ensemble classifier under two-level romantic and relax emotional data.