Accurate and Interpretable Breast Cancer Diagnosis Using Logistic Regression: An Evaluation on the Wisconsin Diagnostic Dataset

Breast cancer remains one of the most prevalent and life-threatening diseases globally, necessitating the development of accurate, stable, and interpretable diagnostic tools. This study aims to evaluate the performance and stability of the classic Logistic Regression model on the standard Wisconsin Diagnostic Breast Cancer (WDBC) dataset, with emphasis on clinically important metrics such as sensitivity and specificity. This research was conducted on the public WDBC dataset containing 569 samples with 30 cytomorphological features. The Logistic Regression model was implemented within a robust evaluation framework comprising 50 independent runs with different random splits of the data into training and testing sets. Data preprocessing included feature standardization (zero mean and unit variance), but any feature selection or dimensionality reduction was avoided to preserve model interpretability. Model performance was assessed using accuracy, sensitivity (also known as recall), specificity, and the Area Under the Receiver Operating Characteristic Curve (AUROC). The Logistic Regression model achieved a mean accuracy of 93.8%. More importantly, the model demonstrated a sensitivity of 97.2% in identifying malignant cases and a specificity of 91.2% in identifying benign cases. The ROC curve analysis yielded an AUROC of 0.9452, indicating excellent discriminative power. Confusion matrix analysis revealed that the model made only 6 errors out of 212 malignant cases (false negatives) and 29 errors out of 357 benign cases (false positives). This error profile indicates the model's conservative behavior, prioritizing high sensitivity. Furthermore, the results showed remarkable stability across all 50 runs (standard deviation less than 1%). The findings demonstrate that Logistic Regression, when implemented with a rigorous evaluation framework and focus on clinical thresholds, can serve as a sensitive, stable, and interpretable baseline model for breast cancer diagnosis. The strong and consistent performance of this classical model makes it a suitable candidate for clinical decision support, particularly in screening environments where minimizing false negatives is paramount.