Machine learning In breast cancer prognosis and prediction

Abstract

In the human body, cancer is a condition that causes cells to proliferate quickly and uncontrolled across the whole body. It has the ability to arise in any of the billions of cells that build up the human body. Human cells generally become divided and turn into new cells as the requirement for human body. When cells get harmed or turn aged, they perish, and young vesicle replace them. Cancer can take many forms. Cancer is normally designated after the limb or tissues in which it arises. For instance, kidney cancer starts in the kidney, blood cancer starts in the blood cells and breast cancer starts in the human breasts. Cancer in breast is the maximal prevalent and frequent disease in female population all over the world. The majority of women identified with breast cancers are just above 50 in age, but breast cancer may strike anybody at any age. In the developed world, in one out of every eight women is diagnosed with breast cancer. However, early detection can help to prevent deaths and save many lives. This paper focuses on prediction and prognosis of cancer in breast using ML models where the paper provides accuracy of the ML deep learning models in diagnostically identifying 569 patients where 212 malignant and 357 benign Fine Needle Aspirate ( FNAs) and its potential accuracy. Also, Recall and the feature numbers in the database is obtained, which is depicted visually. First of all, we have given an overview of ML and deep learning approaches including DT, KNN and Linear SVC and ANN. We examine their BC implications. The Wisconsin breast cancer database (WBCD) is a standard database for assessing results using multiple techniques. This data set shows features such as tumor radius, concavity, texture and fractal dimensions also defined the tumor as Benign or Malignant. After implementing our selected models we find out the most efficient model with respect to precision, recall, F1 score accuracy and confusion metric. We observed that ANN obtains the height accuracy, which is 97.9%. We provided the necessary statistics and graphs in our result part in this paper. We believe that our results may assist lead to more accurate and guided screening in the future.