Effective feature extraction and selection are crucial for the accurate classification and prediction of hand gestures based on electromyographic signals. In this paper, we systematically compare six filter and wrapper feature evaluation methods and investigate their respective impacts on the accuracy of gesture recognition. The investigation is based on several benchmark datasets and one real hand gesture dataset, including 15 hand force exercises collected from 14 healthy subjects using eight commercial sEMG sensors. A total of 37 time- and frequency-domain features were extracted from each sEMG channel. The benchmark dataset revealed that the minimum Redundancy Maximum Relevance (mRMR) feature evaluation method had the poorest performance, resulting in a decrease in classification accuracy. However, the RFE method demonstrated the potential to enhance classification accuracy across most of the datasets. It selected a feature subset comprising 65 features, which led to an accuracy of 97.14%. The Mutual Information (MI) method selected 200 features to reach an accuracy of 97.38%. The Feature Importance (FI) method reached a higher accuracy of 97.62% but selected 140 features. Further investigations have shown that selecting 65 and 75 features with the RFE methods led to an identical accuracy of 97.14%. A thorough examination of the selected features revealed the potential for three additional features from three specific sensors to enhance the classification accuracy to 97.38%. These results highlight the significance of employing an appropriate feature selection method to significantly reduce the number of necessary features while maintaining classification accuracy. They also underscore the necessity for further analysis and refinement to achieve optimal solutions.

Surface electromyography (sEMG)-based hand gesture recognition systems often experience significant performance degradation when deployed in heterogeneous populations, particularly among individuals with lifestyle-related physiological variations such as smoking or alcohol consumption. These variations induce distributional shifts that conventional models fail to address, reducing their practical applicability. To overcome this challenge, we introduce an adaptive incremental k-Nearest Neighbors (ADINC-kNN) algorithm designed to maintain robust classification performance without requiring full model retraining. The proposed method integrates a sliding-window buffer with distance-weighted voting to dynamically refine decision boundaries, enabling smooth adaptation from baseline populations (non-smokers/non-drinkers) to target groups (smokers/alcohol consumers). Extensive evaluation was conducted on 15 hand force exercises performed by 14 subjects using 5-fold cross-validation. ADINC-kNN consistently outperformed static kNN, achieving substantial improvements in accuracy, precision, recall, and F1-score across target populations, with classification performance exceeding 90 % in both smoking and alcohol groups. Although the prediction time per fold increased, the algorithm achieved a superior balance between computational efficiency and predictive accuracy compared with state-of-the-art approaches that typically rely on costly retraining. These results demonstrate the effectiveness of ADINC-kNN as a scalable and practical solution for robust sEMG-based gesture recognition. Its ability to adapt dynamically to population-specific physiological variations makes it particularly suitable for real-world applications in rehabilitation, assistive technology, and human–machine interaction, where user diversity and changing conditions are unavoidable.