DEVELOPING PREDICTIVE MODELS AND PERSONALIZED TREATMENT PLANS FOR COPD EXACERBATIONS USING DATA SCIENCE TECHNIQUES

Abstract

COPD is an ongoing lung disease that significantly adds to both morbidity and mortality all over the world. When lung disease gets worse quickly, it can result in hospital stays, a lower standard of living and greater health costs. It is still hard to forecast these increases in symptoms because different patients react differently and older scoring methods are insufficient. This study uses machine learning methods to predict COPD exacerbations, filling a gap in how accurate predictions and personal care can be. Many models find it hard to be used widely because they don’t properly account for relationships among different types of factors. To resolve the issue, the study designs a framework that runs Logistic Regression, Random Forest and Gradient Boosting Machine (GBM) algorithms on a chronic disease dataset. Its accuracy reached 91%, ROC-AUC hit 0.95 and log loss was only 0.22, surpassing Random Forest (89%) and Logistic Regression (85%). The results indicate that boosting techniques are effective at dealing with non-linear relationships and increasing how well models predict future events. The research team also uses risk stratification to put patients in low-, medium- or high-risk groups to guide personalized treatment suggestions. This research suggests that machine learning, specifically GBM, helps improve early detection and specially tailored care for those with COPD. Issues such as trusting just a single dataset and dealing with how much computing power is needed also exist. Real-time sensor bringing together, checking the data used and making the models clearer for clinical use will be the main topics of upcoming development