Gaussian process latent variable models-ANN based method for automatic features selection and dimensionality...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electromyography (EMG) signals are crucial for prosthetics, rehabilitation, and human-computer interfaces.
• High dimensionality of EMG features complicates accurate classification and increases computational complexity.
• Existing methods struggle to balance classification accuracy with computational efficiency.

Purpose of the Study:

• To introduce a novel approach integrating feature reduction with artificial neural networks (ANNs) for high-dimensional EMG classification.
• To enhance EMG classification accuracy while substantially reducing computational costs.
• To explore the impact of various dimensionality reduction techniques on EMG data.

Main Methods:

• Extracted time and frequency domain features from twelve EMG signal channels.
• Applied dimensionality reduction techniques: PCA, LDA, PPCA, Lasso, and GPLVM.
• Classified reduced-dimension features using an Artificial Neural Network (ANN).

Main Results:

• Linear Discriminant Analysis (LDA) was found unsuitable for this dataset.
• Dimensionality reduction did not significantly impact classification accuracy but greatly reduced computational cost.
• Generalized পরিসংখ্যান (GPLVM) offered the shortest computational time (29s), followed by PCA (35s).
• A 5-feature set demonstrated the best performance among tested feature subsets.

Conclusions:

• Dimensionality reduction is effective in improving the accuracy of movement recognition in myoelectric control.
• The proposed integrated approach offers valuable implications for optimizing EMG-related processes.
• This method provides a computationally efficient solution for high-dimensional EMG signal analysis.