A Comparison of the Analysis of Methods for Feature Extraction and Classification in SSVEP BCIs

• Published in: SN computer science Vol. 5; no. 4; p. 356
• Main Authors: Heidari, Hoda, Einalou, Zahra, Dadgostar, Mehrdad, Hosseinzadeh, Hamidreza, Kalhor, Leila
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 01.04.2024; Springer Nature B.V
• Subjects: Accuracy; Amyotrophic lateral sclerosis; Cardiovascular disease; Classification; Computer Imaging; Computer Science; Computer Systems Organization and Communication Networks; Data mining; Data Structures and Information Theory; Decision analysis; Decision trees; Electroencephalography; Entropy (Information theory); Evaluation; Eye movements; Feature extraction; Feature selection; Fourier transforms; Human-computer interface; Information Systems and Communication Service; Kurtosis; Machine learning; Methods; Multilayer perceptrons; Neural networks; Original Research; Pattern Recognition and Graphics; Principal components analysis; Signal processing; Software Engineering/Programming and Operating Systems; Support vector machines; Temporal resolution; Vision; Wavelet transforms; Wavelet; Steady-state visual evoked potential (SSVEP); Principal component analysis (PCA); Classification; Brain–computer interface (BCI)
• ISSN: 2661-8907; 2662-995X; 2661-8907
• DOI: 10.1007/s42979-024-02638-2

Most brain–computer interface (BCI) systems operate on the basis of electroencephalography (EEG) due to their straightforwardness of applications and high temporal resolution of brain signals. One of the most helpful tools in BCI systems is the steady-state visual evoked potential, which is derived from the EEG data. This work evaluates a number of feature extraction-based methods for evaluating Shannon entropy, skewness, kurtosis, mean, and variance. A number of feature selection techniques, including decision trees, principal component analysis (PCA), t tests, and Wilcoxon, are also assessed. Several classification techniques, including the k nearest neighbor, support vector machine, Bayesian classifier, and multilayer perceptron neural network, were contrasted in the decision step. The decision tree and PCA are used to define a relatively new feature selection method that is the base of the present study. Finally, based on the four acquired frequencies as well as which represent the four directions of right, left, up, and down, the greatest percentage of accuracy was 91.39%.