Machine Learning Modelling

The three machine learning algorithms are LASSO logistic regression, XGBoost, and CatBoost. Model training will use standardized predictors, 10-fold cross-validation, and grid-search hyperparameter tuning. Feature selection removes near-zero variance predictors and then uses ensemble importance rankings to define a final feature set. Patient-level splitting is used to reduce overly optimistic estimates from repeated observations on the same patients. The protocol will develop models for asthma, COPD, and combined asthma-COPD cohorts across two outcome types and three time windows.