Leveraging ANN and LDA Classifiers for Characterizing Different Hand Movements Using EMG Signals

The analysis of electromyographic (EMG) signals has expedited the use of a wearable prosthetic arm. To this end, pattern recognition-based myoelectric control schemes have shown the promising results; however, the choice of classifier and optimal features is always challenging. This paper presents the comparative analysis of classifiers for multiple EMG datasets including (1) the publicly accessible NinaPro database which provides data recorded for 52 hand movements collected from 27 subjects out of which twelve finger movements were classified, and (2) the data collected from ten healthy and six amputee subjects for 11 different hand movements. The classification results of artificial neural networks (ANN) were compared with those of linear discriminant analysis (LDA) for both datasets separately. For dataset 1, the mean classification accuracy of LDA obtained was 85.41% while ANN showed 91.14% accuracy. Similarly, for dataset 2, the mean classification accuracy achieved with LDA was 93.54% while with ANN, it was 97.69%. Besides, p-values were determined for both datasets which revealed better classification results of ANN as compared to LDA. The overall results of this study show that ANN performed better classification and recognition of hand movements as compared to LDA. The findings of this study offer important insights regarding the selection of classifiers of EMG signals which are critical to evaluating the accurate performance of prosthetic human organs.