Data Balancing in Breast Cancer Recognition Using the Wisconsin Breast Cancer Dataset

Breast cancer is ranked as the most diagnosed cancer among women around the world, and it is one of the main causes of death in women. Accurate and early detection of breast cancer is essential for improving patient outcomes, as it allows for timely treatment and better chances of survival. One major challenge for reliable detection is that the datasets used in medical diagnostics are often imbalanced, with far fewer malignant cases compared to benign ones, which may skew machine learning (ML) methods to yield inaccurate detections. This study explored various ways to use suitable ML classifiers such as Logistic Regression (LR), Random Forest (RF), Multilayer Perceptron Neural Network (MLP NN), Support Vector Machine (SVM), and Decision Tree (DT). Monte Carlo simulations were used to evaluate the recognition performance despite the challenge of data imbalance. The Wisconsin Breast Cancer Dataset (WBCD) is a benchmark database for ML research, with 212 malignant (cancerous) and 357 benign (non-cancerous) cases. From WBCD, an imbalanced dataset was randomly created to simulate real-world diagnostic conditions, and the results are compared with the "truth" when more minority cases were available. To address class imbalance, data balancing techniques such as SMOTE (Synthetic Minority Oversampling TEchnique) and weighted bootstrapping (with 30% and 50% weighting) were applied. The performance of the classifiers was then evaluated on both the imbalanced dataset and a separate cross-validation dataset. Additionally, each classifier’s performance was compared against its own baseline on balanced data to evaluate the impact of the balancing techniques. Among all classifiers, Logistic Regression consistently achieved the best performance, maintaining over 99% accuracy and up to 99.34% recall on the imbalanced dataset, and achieving up to 95.77% accuracy and 93.12% recall on the cross-validation dataset. This makes LR the most suitable classifier for preliminary breast cancer recognition.