EMG Viability Classifier
Python - Scikit-learn - NumPy - Pandas - Signal Processing
An ML pipeline that classifies prosthetic trial candidate viability from EMG sensor data, built for Alberta BIONIX.
Designed a full signal-processing and ML pipeline for the SIAT-LLMD dataset (40 subjects, 9 channels, 1920 Hz) to classify candidate viability for prosthetic knee trials.
Built preprocessing in Python: bandpass (20-450 Hz) and 60 Hz notch filtering, DC removal, and channel selection isolating the four thigh muscles above the amputation site.
Engineered 30 features per window — Du's six time-domain features (integrated EMG, variance, waveform length, zero crossings, slope sign changes, Willison amplitude) across four channels, plus six inter-channel Pearson correlations.
Trained three mode-specific one-class SVMs (sitting, standing, walking) on the able-bodied population, with a weighted scoring system (0.2 / 0.3 / 0.5) producing a final 0-100 viability score.