A Hybrid Approach for Improving the Classification performance of Imbalanced Breast Cancer data
Uday Pratap Singh, Mridu Sahu · 2023
Breast cancer is the most common and significant cause of disease worldwide in women. The treatment of breast cancer requires lots of resources, various breast tumors affect different humans. In recent years researchers have proposed different models for the early prediction of breast cancer. The types of breast tumors are benign and malignant; the benign is innocent and does not harm that much, while the malignant is a cancerous tumor, which is the main reason for cancer in the breast region. So, the classification of the two tumors is very important various pretrained and hybrid model is already used for classification such as CNN (Convolution Neural Network)-based classification, CAD (computer-aided diagnosis), and support vector machine classification. In this paper, we have taken BreakHis dataset and balanced both the classes of tumors cell (number of images benign and malignant) using the data augmentation technique GAN (Generative adversarial network) then applied various pretrained models such as VGG 16, RESNET-50, INCEPTION V3, MOBILE NET V2, DENSENET-201 and achieve the classification accuracy of 92, 87, 90, 79, 92 percent respectively then feature extraction is performed on the data using resnet 50. The accuracy is improved drastically and observed as 96.3%.