Enhancing Breast Cancer Diagnosis Interpretability: Logistic Regression Performance on the Wisconsin Dataset

Abstract

Breast cancer (BC) is one of the deadliest diseases in the world, necessitating the development of high-quality diagnostic systems. Therefore, the present research concentrates on examining the efficiency and reliability of Logistic Regression (LR) in terms of results about clinically important features such as specificity and sensitivity relying on the Wisconsin Diagnostic Breast Cancer (WDBC) Database. This set of open-access data includes 569 cases characterized by 30 cytomorphological features. The classification system was established by conducting a careful experimental design with 50 independent trials and a set of random samples taken during the training and testing procedure. Data preparation included feature normalization in accordance with the principle of zero mean and unit variance, while feature selection and transformation techniques were not employed to make the findings more interpretable. The performance of the models included the measurement of their efficiency, including accuracy, sensitivity, Area Under the Receiver Operating Characteristic Curve (AUROC) and specificity. The Logistic Regression methodology achieved an accuracy of 93.8%.

It is important to note that the model recorded a specificity level of 91.2% for benign cases, while the sensitivity for malignant cases was as high as 97.2%. The performance of the model was further confirmed by the AUROC value of 0.9452 obtained from Receiver Operating Characteristic (ROC) curve analysis. In addition, the results of the confusion matrix analysis indicate that the number of false negatives is confined to six cases for a total of 212 malignant cases and that false positives are equal to 29 of 357 benign cases, giving the impression that the model is conservative and favors sensitivity.

In addition, 50 repetitions of the run produced very stable outcomes of under 1%. All of this suggests that LR is a sensitive, robust, and explicable baseline model for BC diagnosis, provided that it is implemented within a proper evaluation framework and with consideration for clinical settings. Due to its proven reliability, the model that has been tested and validated is well-suited to assist in clinical decision-making.