Data Driven Calssification of Hand Postures for Myoelectric Control

Mtech Thesis Project – IIT Guwahati.
Website and Dataset used

Project Overview

This project is my M.Tech thesis work, focused on developing a data-driven framework for classifying hand gestures using surface Electromyography (sEMG) signals.
The long-term goal is to enable intelligent and intuitive control of hand exoskeletons, which can aid rehabilitation and assistive technologies.

---

Objectives

• Preprocessed raw EMG signals (amplification, filtering, segmentation).
• Extract time, frequency, and statistical features to create structured feature matrices.
• Apply machine learning algorithms (Random Forest, SVM, Extra Trees) for gesture classification.
• Implement Non-Negative Matrix Factorization (NMF) to uncover muscle synergies and reduce dimensionality.
• Achieve high classification accuracy and provide insights.

---

Tools & Technologies

• Languages: Python (NumPy, Pandas, SciPy, scikit-learn, Matplotlib)
• Techniques: Feature Engineering, Signal Processing, Dimensionality Reduction, Machine Learning
• Models: Random Forest, K-nearest neighbours, Extra Trees Classifier, Non-Negative Matrix Factorization (NMF)

---

Key Results

• Processed and analyzed multi-channel EMG data from 8 muscles, 15 gestures, 8 subjects.
• Achieved 98% classification accuracy using Random Forest, SVM, and Extra Trees.
• Extracted 2 dominant muscle synergies using NMF with Variance Accounted For (VAF) > 95%.
• Built a robust, reproducible pipeline from raw EMG signals to final gesture classification.