Multi-disease Prediction Using Hybrid Ensemble Learning and Explainable AI

Chronic health conditions are an important focus in the management of older adult health, including heart disease, hypertension, stroke, diabetes mostly appears jointly caused by common biological mechanisms. Early diagnosis is important for diseases before they occur and minimizing the strain on patients and healthcare infrastructure. Historically, separate models were required to predict each disease, which consumed much more computational resources and efforts. In this study we utilized the MIMIC-IV dataset and proposed a hybrid ensemble learning technique with explainable AI. This architecture brings together seven base classifiers using out-of-fold (OOF) stacking and utilizes CatBoost model as meta-model. Feature selection narrowed the dataset down to most relevant 28 attributes clinically meaningful and for global and local explanations we incorporated SHAP and LIME, Making predictions clear and understandable. The model forecasts the risk of heart disease, stroke, hypertension, and diabetes simultaneously, achieving best performance with AUCs of 0.86 for heart disease, 0.81 for stroke, 0.81 for hypertension, and 0.87 for diabetes. SHAP analysis identified key risk factors aligned with medical expertise, whereas LIME supplied patient-level interpretations. this strategy mitigates the weaknesses of conventional single-disease approaches by identifying interdependencies among diseases and enhancing accuracy and improving model transparency. this framework reveals promise for practical use in early risk identification, healthcare decision support and bias identification, presenting a stable and open tool for medical use.