Novel feature extraction method for signal analysis based on independent component analysis and wavelet transform

Area of Science:

• Signal Processing
• Biomedical Engineering
• Machine Learning

Background:

• Feature extraction is crucial for data processing tasks like classification and clustering.
• Current methods, often based on convolutional neural networks (CNNs), demand significant computational resources and large datasets.
• There is a need for efficient feature extraction methods with lower inference costs.

Purpose of the Study:

• To introduce a novel feature extraction method combining wavelet transform and Independent Component Analysis (ICA).
• To achieve high performance in Electrocardiography (ECG) heartbeat classification with low computational inference cost.
• To evaluate the proposed method's effectiveness using the MIT-BIH database.

Main Methods:

• The proposed method integrates wavelet transform into the ICA mixing matrix for enhanced feature representation.
• Electrocardiography (ECG) heartbeat classification was performed using the MIT-BIH database with four classes: Normal, Vestibular ectopic beats, Ventricular ectopic beats, and Fusion beats.
• Experiments involved various wavelet functions and classifiers, with the best performing model selected via 5-fold cross-validation and Wilcoxon test.

Main Results:

• The proposed feature extraction method combined with a multi-layer perceptron classifier achieved a 95.81% Balanced Accuracy (BAC) score.
• The approach demonstrated superior performance compared to most existing feature extraction methods, excluding CNNs.
• Performance was comparable to CNNs for classes with limited training data, indicating efficiency for smaller datasets.

Conclusions:

• The novel wavelet transform-based feature extraction method offers a competitive alternative to CNNs, particularly in resource-constrained environments.
• The method provides a good balance between high performance and low inference cost, facilitating deployment on devices with limited computing power.
• This approach enhances ECG analysis by offering efficient and effective feature extraction for heartbeat classification.