Alzheimer's disease (AD) risk prediction using machine learning and polygenic risk scores (PRS). We integrated multi-trait PRS (AD and non-AD traits) with demographic and genetic covariates (age, sex, APOE ε4, population structure) across five ML models (CatBoost, XGBoost, Random Forest, Logistic Regression, DNN). Feature Set 10 (PRS + covariates) achieved AUC 0.818 with CatBoost, significantly outperforming PRS-only models (AUC 0.659). SHAP analysis identified APOE ε4 and age as top predictors, with non-AD PRS features contributing meaningfully to risk stratification.