Application of Data Mining Techniques to Predict Breast Cancer

Abstract

Breast Cancer is one of the diseases that causes a higher number of deaths in a year. Amongst woman, Breast Cancer is the second highest disease that causes death, and in Canada, it is a leading cause of death. Early detection of breast cancer makes it most curable cancer in among other types of cancer, early detection and accurate examination for breast cancer ensures an extended survival rate of the patients. Data mining techniques have a growing reputation in the medical field because of high diagnostic capability and useful classification. Machine Learning methods can help practitioners to develop tools that allow detecting the early stages of breast cancer.The objective of this study is to predict breast cancer by using k-nearest neighbor (KNN), Support Vector Machine (SVM), Random forest (RF). Moreover, we conducted in details comparison between the three methods. All the methods can be used alone or with ensemble learning to build a more sophisticated classifier. We use the Wisconsin breast cancer dataset to train and validate all classifiers. Then the performance matrix, i.e., accuracy, recall, precision, are measured at different training and testing dataset. The ensemble learning method based on the maximum voting shows the highest accuracy (98.9 %) compared to other classification techniques.