Machine learning implementation for enhanced respiratory sound classification: Assessing algorithmic efficacy

<bold>Background:</bold> Previous research explored artificial intelligence algorithms to improve respiratory disease identification, reducing observer subjectivity and variability in traditional auscultation. <bold>Objective:</bold> Evaluate the effectiveness of machine learning algorithms in classifying pathological respiratory sounds, including medical diagnoses (COPD, Upper Respiratory Tract Infection, Pneumonia, Bronchiectasis, Asthma, Bronchiolitis or Healthy), sex, and age group. <bold>Methods:</bold> The International Conference on Biomedical and Health Informatics (ICBHI) database was utilized, comprising 920 respiratory sound recordings from 126 subjects. Mel-frequency Cepstral Coefficients (MFCC) and Wavelet techniques were applied for analysis. Various machine learning models, such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Logistic Regression (LR), and Decision Tree (DT), were implemented. Internal evaluation metrics included Accuracy, Cross-Validation Accuracy, Average Precision, Average Recall, Average F1 Score, Cohen's Kappa, and area under the Receiver Operating Characteristic Curve. <bold>Results:</bold> KNN achieved 99% accuracy in diagnoses, and LR attained 99% precision in sex detection. Cross-validation confirmed robustness with <5% standard deviation. Other combinations, such as age group, wheezes, and crackles, also scored 98%-99%. <bold>Conclusions:</bold> The algorithms demonstrated high reliability in classifying pathological sounds, diagnoses, sex, and age group. External validation via cross-validation ensured result independence. Models best fitting external data were selected with an average precision standard deviation of 0.012, notably 0.006 for diagnosis.