Design and Implementation of a Time–Frequency Metaheuristic Framework for Automated Limb-Movement Classification

Reliable limb-movement classification using surface electromyography (sEMG) remains challenging due to signal non-stationarity, inter-subject variability, and high feature redundancy. While prior studies have emphasized classifier design, comparatively less attention has been paid to the systematic formulation of robust experimental methodologies that ensure generalization and feature stability. This paper presents a structured research design and experimental framework for sEMG-based limb-movement recognition, focusing on time–frequency feature representation and large-scale metaheuristic feature selection.

The proposed framework integrates standardized preprocessing, time–frequency decomposition using STFT, EWT, and TQWT, and comprehensive feature extraction encompassing time-domain, spectral, and hybrid descriptors. Feature selection is performed using a diverse set of 44 metaheuristic optimization algorithms, with a composite fitness function balancing classification accuracy and feature compactness. Ensemble-based stability analysis is employed to identify consistently informative features across optimizers and validation folds. Classification performance is evaluated using multiple machine-learning models under both within-subject and cross-subject validation schemes, including Leave-One-Subject-Out testing.

The experimental design is validated on benchmark sEMG datasets, with expected outcomes including substantial feature reduction, improved classification performance, and enhanced generalization. The proposed methodology establishes a reproducible and extensible foundation for robust EMG-based movement recognition.