Performance Evaluation of Classification Models for Respiratory Disease Detection Using Wiener-Ville Distribution Features

The respiratory rate is a vital physiological metric for evaluating an individual's health, and its precise measurement is critical in clinical diagnostics. Photoplethysmography (PPG) signals offer a non-invasive and economical method for assessing respiratory activity. Nevertheless, owing to their vulnerability to noise and motion distortions, effective feature extraction and classification methods are essential for dependable respiratory detection. This study utilizes Wiener-Ville Distribution (WVD) for feature extraction, capturing the time-frequency attributes of respiratory-induced fluctuations in PPG data. A variety of classification methodologies, such as Principal Component Analysis (PCA), Expectation-Maximization (EM), Bayesian Linear Discriminant Classifier (BLDC), EM-PCA, and Harmonic Search (HS), are assessed for their efficacy. Experimental results indicate that the Harmonic Search (HS) classifier attains superior accuracy, surpassing alternative approaches through the efficient optimization of feature selection and classifier parameters. The suggested method underscores the efficacy of HS as the optimal classifier gives prominent accuracy of 95.12% and a minimum error rate of 4.88%.