Low-Cost COPD Screening from Respiratory Sounds Using MFCCs with k-NN, SVM and Decision Trees

Abstract

The aim of our research is to develop a simple and inexpensive solution to detect Chronic Obstructive Pulmonary Disease (COPD) in places where sophisticated medical equipment or specialized medical personnel are not available. COPD is a long-term respiratory disorder that impacts a large portion of the global population. Early detection is crucial for effective management; however, conventional diagnostic methods are often complex, time-intensive, and expensive. This study seeks to simplify the process by introducing an accessible and non-invasive screening approach. We propose using an electronic stethoscope to listen and record the patient's respiratory sound and an artificial intelligence algorithm that automatically assists in detecting this condition. We combined two public lung-sound datasets (121 patients; 926 five-second segments) and extracted 20-dimensional MFCCs per frame. Using a patient-wise 70/30 split, k-NN achieved 81.65% segment-level accuracy, outperforming SVM and Decision Tree. Results support a low-cost screening aid for COPD in resource-limited settings.  Machine learning models show promising results, indicating their ability to support COPD screening and assist clinical decision-making.