Breast cancer causes more deaths in women than any other type of cancer. Mammography is the primary screening test for breast cancer. Medical data from CT scans, PET scans, and MRIs are among the most widely used types of information. The use of data mining techniques has become essential for efficient and precise cancer prediction and detection since the work of analysing this massive amount of data has gotten increasingly difficult. Clinically relevant information can be mined from medical photographs to better aid in illness diagnosis and early detection, which is the primary focus of medical image mining. Patients need careful symptom observation and a prediction automatic system that can identify the tumour as benign or malignant in order to receive effective treatment. While its primary function as a generic convolutional neural network is to classify images where the input is an image and the output is a single label, in biomedical applications it also allows us to detect the presence of disease and pinpoint its exact location. This issue can be fixed using deep learning techniques. For the purpose of segmenting and prediction of tumour zone in mammography images, a Grad cam with soft-UNET based architecture is proposed. In order to determine the stage of breast cancer, the network relies on a fully convolutional network that has been upgraded and expanded in design to function with less training images and to provide more accurate segmentations with tumour height and width