Two phases based training method for designing codewords for a set of perceptrons with each perceptron having...

Area of Science:

• Machine Learning
• Artificial Neural Networks
• Biomedical Signal Processing

Background:

• Accurate classification of cardiac arrhythmias is crucial for patient diagnosis and treatment.
• Conventional perceptron models with sign activation functions have limitations in handling complex data patterns.
• The need for efficient and error-free classification methods in medical diagnostics is paramount.

Purpose of the Study:

• To propose a novel two-phase training method for designing perceptrons with multi-pulse activation functions.
• To apply this method for classifying two types of tachycardias.
• To demonstrate the efficiency and accuracy of the proposed method compared to conventional approaches.

Main Methods:

• Initializing perceptrons based on input feature vector dimensions.
• Designing single-pulse activation function perceptrons.
• Developing multi-pulse activation function perceptrons from single-pulse ones.
• Assigning codewords based on multi-pulse perceptron outputs and checking conditions.

Main Results:

• The proposed method successfully classifies two types of tachycardias.
• Achieves zero classification error for linearly separable feature spaces.
• Demonstrates superior performance over conventional perceptrons with sign activation functions in numerical simulations.

Conclusions:

• The two-phase training method effectively designs perceptrons with multi-pulse activation functions for accurate classification.
• This approach guarantees no classification error when data is linearly partitionable.
• The proposed method offers a significant advancement over traditional perceptron models in arrhythmia classification.