Lung Sense-AI: An Interpretable Machine Learning Framework for Respiratory Sound Classification Using Acoustic Feature Engineering

Abstract

According to recent studies by the World Health Organization (WHO), over 262 million individuals suffer from asthma and approximately 3.23 million deaths are attributed annually to COPD. Manual diagnosis based on auscultation and symptom evaluation often leads to misclassification due to overlapping sound patterns and human subjectivity, resulting in delayed or improper treatment. To overcome these challenges, this research proposes an automated Machine Learning (ML) framework for classifying respiratory diseases such as Asthma, Bronchial conditions, COPD, Pneumonia, and Healthy subjects using the International Conference on Biomedical and Health Informatics (ICBHI) respiratory sound dataset. The proposed system begins Mel-Frequency Cepstral Coefficient (MFCC) feature extraction to capture frequency-domain and temporal variations in breathing cycles. These features are then utilized to train and evaluate multiple ML classifiers for comparative analysis including Support Vector Machine (SVM), K-Nearest neighbours (KNN), Decision Tree Classifier (DTC), Adaptive Boosting (AdaBoost), and Linear Discriminant Analysis (LDA). Building upon the limitations of these existing approaches, the proposed Light Gradient Boosting Machine (LGBM) classifier efficiently handles high-dimensional data, improves classification accuracy, and reduces computational complexity through gradient boosting-based ensemble learning. The model outputs disease labels: ["Asthma", "Bronchial", "COPD", "Healthy", "Pneumonia"]. Experimental evaluation demonstrates that the proposed LGBM classifier achieves superior accuracy and faster convergence compared to traditional models, establishing a scalable and reliable approach for real-time respiratory disease detection using non-invasive acoustic analysis. Furthermore, the system supports early diagnosis, reduces dependency on clinical expertise, and enables integration into smart healthcare systems for efficient patient monitoring.