ENHANCED COPD RISK PREDICTION USING CLINICAL DATA: A MACHINE LEARNING-BASED COPD-RINET FRAMEWORK

Abstract

Chronic Obstructive Pulmonary Disease (COPD) is a long-term lung disease that makes normal breathing more difficult as it develops, often underdiagnosed in its early stages, particularly in settings lacking spirometry. This study presents COPD-RiNet, a machine learning (ML) framework for early COPD detection using non-invasive clinical and hematological features. The framework utilizes data cleaning and reduces dimensionality by using principal component analysis (PCA). For classifying the results, it uses a combined approach that includes Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF) algorithms. The model will be assessed using a unified clinical dataset, where it is portioned between 70%for training and 30% for testing. The performance will be further examined using a 10-fold validation approach.COPD-RiNet achieved 99.2% accuracy, 0.98 F1-score, and 0.97 AUC-ROC, surpassing models such as FDDLM, COPD-MMDDxNet, and Dragonfly-Optimized KELM. Leveraging routinely collected parameters, COPD-RiNet offers a feasible solution for early diagnosis in primary care and resource-limited settings, with future work aimed at incorporating disease severity assessment and progression prediction.