EMG features dataset for arm activity recognition.

This study presents a dataset on hand gesture recognition using electromyography (EMG) signals. The data was collected from eight healthy subjects aged between 19 and 35 years, with each subject performing three distinct hand gestures (lifting, grabbing, and flexing). Surface EMG signals were recorded using the Delsys Trigno Wireless biofeedback system from four sensors placed on the dominant hand's Palm A, Palm B, Biceps, and Forearm. The signals were sampled at 2000 Hz and segmented into gesture trials for analysis. The raw EMG data were filtered and processed to extract seven time-domain features across each channel, resulting in 28 total features. These features were reduced using Principal Component Analysis (PCA) to six components, which accounted for 95 % of the variance. The dataset was then used to train and test machine learning models (Random Forest and Logistic Regression) for gesture classification. This dataset has potential reuse in developing gesture recognition algorithms, enhancing prosthetic control, or exploring human-computer interaction (HCI) applications.