Social Sciences · Urban Studies
Comprehensive Analysis of Feature Extraction Methods for EEG-based Emotion Classification
S Sivaranjani, V Prithiga, S Ragavi, S Reena, Sanchana Sri S
See on the mapAbstract
Emotion recognition from EEG signals is widely applicable and a rapidly emerging field in human-computer interaction, mental health analysis, and affective computing. The main problem of addressing a novel enhanced method for EEG-based emotion classification using the Discrete Wavelet Transform (DWT) to extract its features into delta, theta, alpha, beta, and gamma frequency components, and essential time-frequency features could be extracted that can improve the accuracy of the classification. The features are fed into the machine learning classifiers, namely CNN, RNN, SVM, LDA, and Decision Trees, in order to obtain a classification accuracy of 96%. The experimental results indicate a progressive increase in accuracy and consistent reduction in loss over 20 epochs, showing the effectiveness of DWT-based feature extraction. The study evaluates model robustness by using key performance metrics like accuracy, sensitivity, and specificity. It shall improve in future, including optimized wavelet parameters, refinement in architectures of the neural network, real-time processing improvement, and multimodal data fusion to be incorporated for enhancement of system performance.