Breast Cancer Diagnosis: Hypothesis Testing Logistic Regression vs XGBoost/SVM/RF (F1=0.981, Wisconsin Dataset)

Background: Breast cancer caused 2.3M cases & 670K deaths (2022, WHO). FNA cytology accuracy varies 72-91.6% due to pathologist variability. Early detection >99% 5-year survival demands high malignant recall.Methods: Tested LR/SVM/RF/XGB on Wisconsin dataset (n=569, 30 features, 37% malignant). StandardScaler + 10-fold CV GridSearchCV (F1-macro). Paired t-test (df=9, α=0.05) vs H₀: ensembles=no superiority.Results: LR best F1-macro 0.981±0.028, AUC 0.995±0.013, test recall 97.6% (41/42 TP). t(9)=-0.67, p=0.520 (fail H₀). Worst radius/perimeter top features.Files: Preprint PDF (11p, APA), reproducible .ipynb, modelresults.csv, ablation_results.csv, 8x 300DPI figures (confusion matrix, XGBoost importance, PCA, pipeline). random_state=42, scikit-learn 1.6, requirements.txt. Code: github.com/VivianMaduekwe/breast-cancer-ml-preprintLicense: CC BY 4.0