Breast cancer is increasingly common in women under 50, and outcomes vary sharply by subtype, demographics, and treatment regimen, making tailored prognosis essential for personalized care. As a team, we explore how five-year survival differs across breast cancer subtypes and whether it can be reliably predicted from routine clinical and genomic information. Our analysis uses the METABRIC cohort from cBioPortal, ~1980 UK and Canada patients with rich clinical, pathological, and copy-number features. We benchmark four models including Logistic Regression, LightGBM, XGBoost, and SVM under cohort-based cross-validation, with MICE imputation applied per cohort to prevent leakage. Beyond predictive accuracy, we quantify the survival gap for triple-negative (Basal/Claudin-low) patients and identify which clinical and genomic features drive the model's decisions.