Transfer Learning with XGBoost for Predictive Modeling in Electronic Health Records

The integration of XGBoost and transfer learning for modeling predictions in Electronic Health Records (EHR) is investigated in this study. The study uses a deductive methodology, an interpretive philosophy, a descriptive research design, and secondary data collection. The accuracy of prediction, accessibility, and generalization of the XGBoost models enhanced by transfer learning are assessed in a variety of patient populations. The outcomes demonstrate noteworthy enhancements in precision and flexibility, confirming the effectiveness of the suggested methodology. Metrics of interpretability like SHAP values provide information about the level of openness of a model. The models' equilibrium over time is revealed by temporal analysis. The trade-offs between interpretability and accuracy are brought to light by critical analysis, which inspires suggestions for additional research.