Research on GA-SVM Based Head-Motion Classification via Mechanomyography Feature Analysis

Area of Science:

• Biomechanics
• Biomedical Engineering
• Signal Processing

Background:

• Mechanomyography (MMG) signals offer a non-invasive method for analyzing muscle activity.
• Classifying head motions is crucial for applications in human-computer interaction and rehabilitation.
• Accurate and efficient classification of MMG signals remains a challenge.

Purpose of the Study:

• To investigate the classification of six distinct head motion types using MMG signals.
• To evaluate the effectiveness of different feature sets and machine learning models for MMG signal classification.
• To determine optimal parameters for achieving high classification accuracy.

Main Methods:

• MMG signals from head motions were segmented using an unequal segmenting algorithm.
• Three feature types (time domain, time-frequency, nonlinear dynamics) were extracted into five feature sets.
• A genetic algorithm-optimized support vector machine (GA-SVM) with a radial basis function (RBF) kernel was employed for classification.

Main Results:

• The GA-SVM classifier achieved the highest accuracy using the RBF kernel.
• Combinations of three feature sets consistently yielded over 80% average classification accuracy.
• The best performance, reaching 88.2% accuracy, was obtained with feature sets 2, 3, and 5.
• Using four MMG signal channels and at least 60 training samples ensured satisfactory classification accuracy.

Conclusions:

• The GA-SVM approach, particularly with the RBF kernel and combined feature sets, is effective for classifying head motions from MMG signals.
• Feature set selection, number of signal channels, and training sample size significantly impact classification performance.
• This methodology provides a robust framework for developing advanced head motion detection systems.