Deep‐features with Bayesian optimized classifiers for the breast cancer diagnosis

Abstract

AbstractTo improve the state‐of‐the‐art works in breast cancer diagnosis, this paper proposes ensemble‐based deep‐transfer learning with classifiers namely optimizable K‐nearest neighbors, optimizable naïve Bayes, optimizable ensemble, and optimizable support vector machine algorithms for the auto‐feature extraction, detection, and severity classification of input mammograms. The hyperparameters of these classification algorithms are optimized by using the Bayesian optimization technique. The extracted robust features are normalized and then classified using the Bayesian optimized classifiers. Thus, the work throws a light of research on detecting whether the input mammogram is normal or abnormal. Afterward, it further focuses on the severity classification of abnormalities that is, benign or malignant. The aforesaid algorithms are trained and tested for the three‐class classification problem, thus achieving a maximum performance using the Bayesian optimized SVM algorithm applied with ResNet18 deep features providing classification accuracy of 99.689% for MIAS and 98.883% for INbreast dataset.