A Theoretical Framework for Adaptive Neural-Machine Interfaces in Upper Limb Prosthetics: Addressing Signal Non-Stationarity in Nigerian Amputee Populations

Myoelectric upper limb prostheses depend on stable electromyography signal patterns for reliable control, but EMG signal non-stationarity arising from electrode displacement, fatigue, and sweat accumulation under Nigerian tropical conditions creates systematic control degradation that existing adaptive control frameworks developed in temperate climates do not adequately address. This dissertation develops a theoretical framework for adaptive neural-machine interfaces that explicitly models and compensates for EMG non-stationarity sources in upper limb prosthetic control under Nigerian amputee conditions. The framework synthesises concepts from non-stationary time series theory, online machine learning, transfer learning, and biomechanical modeling to construct a multi-layer adaptive control architecture. At the signal acquisition layer, the framework specifies a non-stationary electrode impedance compensation model calibrated to sweat-induced drift patterns characterized empirically in a cohort of forty upper limb amputees at the National Orthopaedic Hospital Lagos over eighteen months. At the feature extraction layer, a co-variate shift detection algorithm triggers re-calibration protocols when distribution shift between current and reference EMG feature space exceeds a theoretically derived threshold. At the classification layer, an incremental support vector machine variant with selective memory is formulated to balance adaptation speed against catastrophic forgetting. The full adaptive framework is implemented in a real-time embedded prosthetic controller and evaluated against a static non-adaptive classifier baseline. Performance evaluation over twelve weeks of daily use by fifteen prosthesis users in Lagos demonstrates that the adaptive framework maintains classification accuracy above 85 percent throughout, compared to static classifier degradation to below 60 percent by week six. Original theoretical contributions include the non-stationarity characterization model and the distribution-shift-triggered adaptation criterion. Keywords: neural-machine interface, myoelectric prosthetics, EMG non-stationarity, adaptive control, Nigerian amputees.