Classification of Upper Arm Movements from EEG signals using Machine Learning with ICA Analysis
This work addresses the problem of accurately classifying upper limb movements for brain-computer interface applications, representing an incremental improvement with specific gains in performance metrics.
The paper tackled the classification of left/right-hand movements from EEG signals using a Multi-layer Perceptron Neural Network with ICA for artifact removal, achieving a combined accuracy of 96.02% and intra-subject accuracy of 94.72% across nine subjects.
The Brain-Computer Interface system is a profoundly developing area of experimentation for Motor activities which plays vital role in decoding cognitive activities. Classification of Cognitive-Motor Imagery activities from EEG signals is a critical task. Hence proposed a unique algorithm for classifying left/right-hand movements by utilizing Multi-layer Perceptron Neural Network. Handcrafted statistical Time domain and Power spectral density frequency domain features were extracted and obtained a combined accuracy of 96.02%. Results were compared with the deep learning framework. In addition to accuracy, Precision, F1-Score, and recall was considered as the performance metrics. The intervention of unwanted signals contaminates the EEG signals which influence the performance of the algorithm. Therefore, a novel approach was approached to remove the artifacts using Independent Components Analysis which boosted the performance. Following the selection of appropriate feature vectors that provided acceptable accuracy. The same method was used on all nine subjects. As a result, intra-subject accuracy was obtained for 9 subjects 94.72%. The results show that the proposed approach would be useful to classify the upper limb movements accurately.