Bioelectric signal classification using a recurrent probabilistic neural network with time-series discriminant...

Area of Science:

• Computational Neuroscience
• Machine Learning
• Signal Processing

Background:

• High-dimensional time-series data present classification challenges.
• Existing methods may lack efficiency or accuracy for complex patterns.
• Probabilistic neural networks offer a framework for uncertainty quantification.

Purpose of the Study:

• To develop a probabilistic neural network for high-dimensional time-series pattern classification.
• To integrate time-series discriminant component analysis (TSDCA) into a neural network framework.
• To improve classification accuracy and reduce computational training time.

Main Methods:

• Developed a probabilistic neural network based on time-series discriminant component analysis (TSDCA).
• Utilized orthogonal transformations for high-dimensional time-series compression into a lower-dimensional space.
• Employed a continuous-density hidden Markov model with a Gaussian mixture model in the reduced space.
• Integrated the analysis into a neural network trained via a backpropagation-through-time algorithm.

Main Results:

• Achieved high-accuracy classification of high-dimensional time-series patterns.
• Demonstrated significant reduction in computation time for network training.
• Validated the proposed network's effectiveness using electroencephalogram (EEG) signals.

Conclusions:

• The proposed TSDCA-based probabilistic neural network effectively classifies high-dimensional time-series data.
• The method offers a computationally efficient approach for complex pattern recognition.
• The network shows promise for applications in neuroscience, particularly with EEG data analysis.