Hybrid Deep Learning and Machine Learning Framework for COPD Severity Classification and Disease Progression Prediction

Abstract

Objectives To design an integrated machine learning and deep learning architecture for accurate chronic obstructive pulmonary disease (COPD) severity classification and disease progression estimation utilizing medical and spirometry data. 
Materials and methods A harmonized COPD medical dataset consisting demographic, physiological, spirometry, and longitudinal patient data was used for classification of disease severity and progression forecasting. The classification of disease severity dataset comprised of 1030 single-visit patient data with several variables. For temporal disease progression evaluation, the same patient attributes were structured into longitudinal sequences includes up to 8 yearly visits per patient, yielding to 8240 temporal observations for deep learning-based progression modelling.  
Results The experimental analysis illustrated that among Random Forest, Gradient Boosting and Support Vector Machine, the Support Vector Machine accomplished the best prediction performance for COPD severity classification with an accuracy of 95.34%. For temporal disease trajectory evaluation, the Bidirectional Long Short-Term Memory (LSTM) model adequately predicted future COPD exacerbation risk and longitudinal respiratory deterioration patterns with accuracy of 95.08%. Further, the developed integrated ensemble architecture combined with regression based and temporal deep learning estimations enhanced the overall forecasting reliability for both COPD GOLD classification and future exacerbation risk trajectory, accomplishing a definitive accuracy of 95.63% and F1-score of 95.62%
Conclusion The developed integrated architecture optimally enhanced classification of COPD severity and future exacerbation forecasting. The combination of machine learning and deep learning models improved prediction accuracy and provide early medical decision-making.
Clinical Relevance Statement The designed architecture provides prior identification of severe COPD cases and high-risk patients with possible deteriorations. This facilitates periodical medical intervention, individualized treatment planning, and enhanced management of respirator related disease.