Breast Cancer Classification Using Ensemble Hard Voting with Random Under-Sampling

Abstract

Breast cancers emerged as one of the fatal diseases among women. Particular efforts are being taken continuously around the globe to help this cause. Taking advantage of Machine Learning (ML) technology is a measure to lessen the death toll of this disease. This paper has emphasized breast cancer classification to expedite the diagnostic process by identifying whether the tumor is benign or malignant. Moreover, in this process, we have combined the ensemble learning-based Hard Voting (HV) technique with random under-sampling using the Wisconsin Diagnostic Breast Cancer (WDBC) dataset. This dataset provides several features upon performing Fine Needle Aspirate (FNA) test to each affected individual, and five features have been chosen to accomplish the classification task. These features have been further scaled using RobustScaler, and the classes (benign or malignant) were balanced using random under-sampling to achieve an accurate outcome. Four distinguished ML classifiers, e.g., Decision Tree Classifier (DTC), k-Nearest Neighbor (KNN), Random Forest (RF), and Support Vector Machine (SVM), have been applied to form an HV meta-classifier. The HV meta-classifier has obtained 99.42% test accuracy in classifying breast cancer incidents. This result has been further verified by considering other distinct parameters.