Classification of EEG Signals from Motor Imagery of Hand Grasp Movement Based on Neural Network Approach

Every human movement is controlled by the brain, that can read in the form of EEG signals. The classification in EEG signals is very difficult, this is because the data is dissimilar. Neural Network has become one of the most dominant ways to increase the classification accuracy of these signals. The purpose of this study is to discover an appropriate combination for the best classification accuracy of right-hand grasp movement based on EEG headset. There is three movement classification: grasping, relaxing, and opening hand. these classifications take profit from event-related desynchronization and event-related synchronization phenomenon that makes it possible to differ relaxation, imagery, and movement state from each other. Determination combinations of electrodes used based on Genetic Algorithm, every combination divide by several groups of the electrode. every signal that has been carried out by filtering process using Independent Component Analysis (ICA), Bandpass filter, and spectrum analysis using Fast Fourier Transform (FFT). Maximum Mu and Beta power with the frequency being features that will be used in classification. Classification uses several neural network algorithms, namely Probabilistic Neural Network, Radial Basis Network, Exact Radial Basis Network, and General Regression Neural Network. The average values of classification accuracy are 53.08% for training, and 50.68% for testing. The best classifier is Probabilistic Neural Network (PNN) with the value of accuracy was reached 61.96%.