Detection of COPD-like Breathing Patterns from Non-Invasive Pressure and Flow Data Using Machine Learning and Deep Learning

Chronic obstructive pulmonary disease (COPD) remains a major public health burden, with early detection being critical to improving outcomes. Traditional diagnostic tools often rely on spirometry, which can be invasive, effort-dependent, and unsuitable for large-scale screening. This study explores the use of non-invasive respiratory pressure and flow signals to classify COPD-like breathing patterns using machine learning (ML) and deep learning (DL) approaches. A dataset of simulated respiratory waveforms was segmented into fixed-length windows, from which statistical and physiological features were extracted. Four traditional ML classifiers, including Random Forest, Logistic Regression, Support Vector Machine, and k-Nearest Neighbors, were evaluated alongside DL architectures such as Convolutional Neural Networks and Long Short-Term Memory networks. Model interpretability was enhanced using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME), which highlighted pressure mean, volume, and pressure variability as key predictive features. These findings demonstrate that integrating signal-derived features with advanced ML and DL models enables accurate and interpretable detection of COPD-like respiratory dynamics. The proposed framework provides a step toward non-invasive, automated screening methods that could complement clinical assessments and support early intervention strategies.