Ensemble learning-based detection of hemiplegia using electromyographic signals

Abstract

Hemiplegia is a medical condition that can have a profound impact on an individual's mobility and gait, frequently stemming from strokes. Physical examination is typically utilized as the primary diagnostic method for this ailment. Although a more thorough evaluation can be attained by utilizing electromyographic signals, precisely capturing muscle activity is difficult due to variability and noise, impairing classification accuracy. This study proposes an ensemble learning approach to overcome these limitations and enhance the accuracy of hemiplegic gait classification. The proposed method leverages the strengths of multiple classifiers to achieve improved classification accuracy. As a comparison, the Random Forest Classifier had the highest accuracy of 97.34%, followed by the decision tree at 96.29% and the k-nearest neighbor at 95.40%. An ensemble of these three classifiers improved accuracy to 98%, accompanied by impressive Precision of 98%, Recall of 97%, F1-Score of 98%, and Area Under the Curve score of 99.70%. The findings of this study can contribute to the development of more accurate and reliable systems for the classification of hemiplegic gait, which can have a significant impact in various clinical and research settings. © 2024 IEEE.