Breast Cancer Diagnosis using Machine Learning Approach

Abstract

Breast cancer is the second most common cancer in women after skin cancer. When cancer care is delayed or inaccessible, there is a lower chance of survival, greater problems associated with treatment and higher costs of care. Early diagnosis improves cancer outcomes and leads to a better prognosis. In third world countries like Nigeria, where state-of-the art breast cancer diagnostic machines and the experts are grossly insufficient, alternative approaches to early diagnosis of breast cancer must be evolved. These preliminary data obtained from images of suspected cases of breast cancer are transformed in profiles of breast diseases, which are used by the local physicians in charge of breast disease patients. Each new case can then be compared by the local treating physician with the profile of all preceded cases with the same diagnosis. Three supervised learning models; Logistic Regression. Random Forest Classifier, and K-Nearest Neighbors were used to train the cancer dataset, and Random Forest Classifier outperformed with accuracy of 96% and an almost perfect sensitivity/Recall index. The dataset could not capture the demographic effects of the breast cancer images on the diagnosis, which now opens up new research areas in this study of breast cancer.