Leveraging Classifier Performance Using Heuristic Optimization for Detecting Cardiovascular Disease from PPG Signals.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Cardiovascular Health

Background:

• Photoplethysmography (PPG) signals measure blood volume changes and are increasingly utilized for non-invasive cardiovascular disease (CVD) detection.
• Analyzing PPG signals aids in identifying irregular heart patterns and other CVD indicators.

Purpose of the Study:

• To evaluate the efficacy of heuristic optimization algorithms combined with machine learning classifiers for non-invasive CVD detection using PPG signals.
• To identify the optimal combination of dimensionality reduction and classification techniques for accurate CVD diagnosis.

Main Methods:

• Utilized PPG data from 41 subjects (21 normal, 20 CVD) from the CapnoBase database.
• Applied heuristic optimization algorithms (ABC-PSO, CSA, DFA) for PPG data dimensionality reduction.
• Classified dimensionally reduced PPG data using various machine learning models (LR, KNN, LDA, SVM variants) and evaluated performance with metrics like accuracy, MCC, and F1 score.

Main Results:

• The SVM-RBF classifier, applied to ABC-PSO dimensionality-reduced data, achieved the highest accuracy (95.12%) and lowest error rate (4.88%).
• This optimal model also yielded high performance metrics: MCC and Kappa (0.90), GDR and F1 score (95%), and Jaccard Index (90.48%).

Conclusions:

• Heuristic-based optimization and machine learning classification of PPG signals demonstrate high effectiveness for non-invasive CVD detection.
• The study confirms the potential of advanced computational methods in improving cardiovascular diagnostics.