This paper presents a machine learning model for determining the stages of breast cancer. The model normalizes images, eliminates noise, and removes artifacts through preprocessing, segmentation, and classification. Patching is used to extract the region of interest (ROI), while multi-thresholding is used for segmentation. The suggested models, including CNN-KNN, VGGNet-16, CNN-naïve Bayes, and CNN-SVM, were tested on a dataset of images from the CBIS-DDSM database. Experiments were conducted using full mammogram images and segmented ROI images, resulting in accuracies of 39%, 85%, 86%, and 91%, respectively. The CNN-SVM model achieved a 6% increase in classification accuracy compared to the state-of-the-art VGGNet-16 model.