Abstract
Breast cancer is one of the most prevalent causes of cancer-related death globally. Preliminary diagnosis of breast cancer increases the patient's chances of survival and healing. In this paper, we propose a hybrid deep transfer learning model integrating xception with support vector classifier (XSV) and xception with random forest (XRF) along with pre-processing technique to classify breast cancer as cancerous (malignant) or non-cancerous (benign) along comparative analysis of prominent machine learning classifiers, such as Random Forest Classifier (RFC), Logistic Regression (LR), Support Vector Classifier (SVC), K-Nearest Neighbors (K-NN), and Ada-boost. In experiment all the models are implemented on two openly accessible datasets: BreakHis and Breast Histopathology Images Database (BHID) across various metrics such as accuracy, area under the receiver operating curve, precision, recall, f1-score, Matthew's correlation coefficient, classification success index, and kappa at different magnification levels of images. Our proposed model that utilized the fine tuning of xception model in conjunction with RFC and SVC, surpass existing breast cancer classification methodologies. Specifically, the XSV that achieved accuracies of 89.26%, 85.87%, 90.17%, and 88.98%, while the XRF attained accuracies of 87.78%, 84.78%, 88.98%, and 87.61% for BreakHis at 40X, 100X, 200X, and 400X magnifications, respectively. For BHID at 40X magnification, the XSV and XRF models achieved accuracies of 87.35% and 87.29%, respectively. Employing this study will aid our medical practitioners and researchers in choosing an accurate model for tumor classification and our results will help medical professionals to classify the disease with precision.