Hybrid Stacking Ensemble With Augmented Features For Robust SSVEP Signal Classification In BCI

Abstract

The Brain Computer Interface (BCI) acts to boost human abilities. Steady-State Visual Evoked Potential signals are one of the popular paradigm used in BCI experiments, where precise classification of these signals is the key for system performance. In this paper, we introduced a meta learner, a hybrid machine learning technique used in the classification of SSVEP signals by utilizing stacking ensemble and augmented features. The proposed methodology uses advanced signal processing methods, such as wavelet denoising, for preprocessing electroencephalogram data and extracting patterns. The classification system incorporates a stacking ensemble model, in which several base classifiers such as k-nearest neighbors, decision trees, and support vector machine (SVM) are trained by the extracted features, and their results are calibrated by a meta-learner in order to enhance the accuracy of the class. The new proposed system is tested on SSVEP signal dataset and proves superior performance when compared with single-model traditional methods. Key performance metrics, such as accuracy, F1 score, precision, specificity, recall, and area under the curve for receiver operating characteristic analysis, are presented in order improve the efficiency of the proposed system. It can be seen from the results that the use of stacking ensemble, along with augmented features, presents a strong solution for SSVEP signal classification, with substantial improvements, with achieved accuracy of 97.13%. This novel solution has the potential to improve the performance of SSVEP BCI application.