Breast Cancer is one of the dangerous diseases which lead in resulting deaths among women. This is due to the presence of cancerous cells that are produced in extra amount of proportion which can replace the neighboring non-cancerous cells or it can infect all over the body. As the breast cancer concerns women mostly at the age of 40, they are asked to attain the regular mammographic screening, since mammography is most reliable method for cancer detection at early stages. Mammogram is the most common method used for breast imaging. It helps in examine the presence of cancer at early stages and help in reducing the mortality rate by 25-30% in screened women. There occur many different types of breast cancer such as: mass, micro calcification clusters, architectural distortion and asymmetry breast tissue. This dissertation carries the masses problem and deals with its shape and texture feature for classification. Various type of techniques and methodologies are present in mammography which helps to find out the presence of cancer and also multiple ways to detect it in its early stage so that the patient affected by it could not lead to death. Mammography is the most common, safe and inexpensive methodology suggested whose standard image database could be used for training the learning machine. In this dissertation nucleus segmentation is used to find out the region of interest (ROI). The result of ROI is further used for extracting the valuable shape and textural features by using geometrical features, GLCM and GLDM for classifying the cancer through the machine learning approach i.e. CNN (Convolution neural networks). CNN remove the overlapping of features obtained after segmentation. Hence, CNN is used to evaluate the performance through defining accuracy, precision, and recall and also compare the results with existing logistic regression and neural network classification technique.