SGHE-BC: A Statistical Guided Hybrid Ensemble Framework for Interpretable Breast Cancer Classification Using Machine Learning

Fine needle aspiration (FNA) cytology for breast cancer diagnosis suffers from interpreter subjectivity and borderline case uncertainty.We created SGHE-BC-a stacking ensemble specifically for the Wisconsin Breast Cancer Dataset 569 samples, 30 morphometric features.Four base models form the core: RBF-SVM, Random Forest, Logistic Regression, and KNN.Each uses tailored preprocessing StandardScaler for SVM, RobustScaler for Logistic, MinMaxScaler for KNN plus sigmoid calibration for trustworthy medical probabilities.Logistic meta-learner stacks predict_proba outputs via 10-fold StratifiedKFold.On 80/20 stratified split, SGHE-BC delivered 97.37% accuracy, 100% precision, 92.86% recall, 96.30% F1, ROC-AUC=0.99.RF feature importance ranked concavity_worst, concave_points_worst highest.Outperforms individual models by 0.88-2.63%accuracy.Calibration ensured reliable risk scores; stratified stacking handled 62.7%/37.3%class imbalance.SGHE-BC could standardize FNA interpretation in resource-limited clinics.