Classification of Imagery Hand Movement Based on Electroencephalogram Signal Using Long-Short Term Memory Network Method

Amputation is sometimes utilized to overcome tissue death in human limbs. Prostheses offer individuals an effective solution for restoring their quality of life. The development of prosthetic control systems using EEG-acquired movement imagery signals is ongoing. This technology has proven a viable option due to its easy controllability by an individual’s thought patterns. This study aimed to discover distinguishing features between imagery movement and grasping and opening hand movements. To this end, the proposed method is a classification using Long-Short Term Memory Network (LSTM) with various feature combinations of mean, standard deviation, variance, RMS, skewness, kurtosis, and PSD at alpha rhythm. Data were acquired from three healthy subjects using the Emotiv Epoc+Headset. The classification results showed that applying skewness and kurtosis features yielded an accuracy range of 73.52% to 100% for each subject’s data. On the other hand, combining kurtosis and Power Spectrum Density (PSD) features resulted in 84.9% accuracy for the subjects’ combined data. This result shows great potential in supporting the development of prosthetic control to improve the quality of life of an amputee.