A Fast and Interpretable Logistic Regression Framework for Breast Tumor Classification Using the Wisconsin Diagnostic Dataset

Early and reliable discrimination between malignant and benign breast tumors is essential for clinical decision-making and for reducing unnecessary invasive procedures. This study presents a lightweight and reproducible machine-learning pipeline that integrates standard feature normalization with logistic regression to classify breast tumors using the Breast Cancer Wisconsin (Diagnostic) dataset (WDBC), which contains 569 samples described by 30 quantitative features derived from digitized fine-needle aspirate (FNA) images [1-3]. We implemented an end-to-end workflow in Python (scikit-learn), including stratified train-test splitting, model training, and evaluation with clinically meaningful metrics such as accuracy, sensitivity, specificity, and ROC-AUC [4-6]. On the held-out test set (n=114), the proposed approach achieved 98.25% accuracy and a ROC-AUC of 0.9954, with a confusion matrix indicating only two misclassifications (1 malignant predicted as benign, 1 benign predicted as malignant). Specifically, when treating malignant cases as the clinically critical positive class, the method yielded 97.62% sensitivity and 98.61% specificity. These results demonstrate that a simple, interpretable model can achieve near state-of-the-art performance on structured biomedical features while remaining computationally efficient and suitable for rapid prototyping.In addition, we benchmarked logistic regression against several classical baselines (SVM with RBF kernel, random forest, and kNN) under the same train-test split and evaluation protocol. The results indicate that increasing model complexity yields limited performance gains on WDBC, suggesting that an interpretable linear classifier can already approach the performance ceiling on this feature-engineered dataset. Future work will focus on external validation, calibration for risk estimation, and multimodal extensions to incorporate imaging or omics signals.