Date of Publication :9th August 2023

Abstract: Motor imagery electroencephalograph is nonlinear, nonstationary and high dimensional in nature. Due to which, the prediction of existing classification model across multiple subjects is limited. To improve the performance of the existing classification models across multiple subjects, a new preprocessing approach is presented in this paper. A hybrid feature selection approach is introduced to select the optimal number of channels followed by clustering. Clustering helps to explore shared brain activity patterns and their relationships to outside factors by detecting similar clusters among different subjects. In this study, four different classifiers are used to classify motor imagery electroencephalograph data. The proposed approach yields an accuracy of 99.6% using ensemble technique. Significant improvement is seen in the Logistic Regression. The results in this study indicate that generalization of motor imagery electroencephalograph across multiple subjects is possible using our proposed approach.

Reference :