Breast cancer detection using ensemble methods

Abstract

Breast cancer (BC) is one of the most common cancers among women. This study's framework is divided into three phases. Firstly, a majority hard voting approach is used to apply an ensemble classification mechanism as a decision fusion technique on the level of convolutional neural networks (CNNs). Five pre-trained CNNs—visual geometry group 19 (VGG19), densely connected convolutional network 201 (DenseNet201), residual network 50 (ResNet50), mobile network version 2 (MobileNetV2), and inception version 3 (InceptionV3)—are evaluated, using a data splitting test ratio represents 30% of the total dataset. Secondly, the classification results of the five CNNs are compared to get the best-performance model. Then, seven state of art machine classifiers—decision tree (DT), histogram-based gradient boosting classifier (HGB), support vector machine (SVM), random forest (RF), logistic regression (LR), gradient boosting (GB), and extreme gradient boosting (XGB)—are used to improve system performance on the feature vector that was taken from this CNN model. Thirdly, to improve robustness, a majority hard voting technique is used at the external classifier level using the highest four classifiers selected based on their accuracy. Several experiments were conducted in this study, and the results showed that ResNet50 produced the best results in terms of precision and accuracy. The majority voting mechanism improves the system’s accuracy to 99.85% through CNNs and to 100% through traditional classifiers.