Respiratory Sound Analysis for Pulmonary Disease Classification using Local Cascade Ensemble

Pulmonary diseases account for approximately 10% of global deaths, with Chronic Obstructive Pulmonary Disease (COPD) alone affecting over 384 million people worldwide. However, traditional diagnostic methods relying on auscultation and radiographic imaging suffer from subjectivity, high misdiagnosis rates, and the need for expert intervention. To address these challenges, this study introduces the ML-Driven Pulmonary Diseases Classification (ML-PDC) framework, leveraging Mel-Frequency Cepstral Coefficients (MFCC) for feature extraction and a Local Cascade Ensemble (LCE) classifier for robust classification. The ICBHI dataset, comprising three classes (Healthy, COPD, and Pneumonia), was used for model training and validation. Unlike conventional deep learning methods that require extensive training data and computational resources, the proposed LCE-based approach integrates multiple weak learners in a hierarchical structure, enhancing decision-making accuracy while preventing overfitting. The ML-PDC model outperforms existing techniques, achieving 98.53% accuracy, 98.63% precision, 98.384% recall, and 98.584% F1 score, making it a reliable, automated tool for pulmonary disease screening and diagnosis.