Performance Based Features for Classification of Cancer Using Images of Breast Mass Through Fine Needle Aspirate

Abstract

Breast cancer is commonly seen in women. types of cancer specially seen in women. It is proved to be the second most reason for death. The early detection of such cancer saves the life of the patients and stops spreading. Even though the country's medical infrastructure has eventually grown, there is a need to address this problem. In search of solution to the problem, this article presents a method to classify breast cancer through fine needle aspiration features into benign and malignant. The data required for the proposed method is collected from Breast Cancer Wisconsin (Diagnostic) dataset. Eight best performing features are identified. Grouping of these features is carried out in a group of two, three, four, five, six, seven and eight to know the behavior of eight classifiers. Classifiers, namely, Decision Tree (DT), K-nearest neighbor (k-NN), Support Vector Machine (SVM), Ensemble Learning (EL), Discriminant Analysis (DA), Logistic Regression (LR), Naïve Bayes (NB), Kernel Approximation (KA) are used in the work. Results reveal that SVM gave an average accuracy of 98.2% with balanced dataset. Results are compared with state-of-the-art methods.