A FFM-based Feature Selection for Breast Cancer Classification: Machine Learning Approach

Abstract

Breast cancer is a major health issue for women worldwide. Early detection of breast cancer can improve
the possibility of curative treatment as well as lower the mortality rate. The aim of this research work is
to build a classification model with relevant features from the {Wisconsin breast cancer dataset (WBCD).
In this research work, we have applied two distinct methodologies to the pre-processed dataset. The first
methodology contains the existing classification model (without feature selection method), while the
second methodology includes the proposed feature selection, i.e. Frequent Feature-set Mining (FFM)
method based on frequent itemset mining concept along with different learning classifiers like Support
Vector Machine (SVM), Naive Bayes, Random Forest, Logistic Regression, Adaptive Boosting
(Adaboost), Decision Tree, and k-Nearest Neighbor (k-NN), and finally evaluated on different statistical
parameters like accuracy, balanced accuracy, sensitivity, specificity, precision, jaccard similarity, dice
coefficient, Receiver Operating Characteristic curve (ROC-Curve) and precision-recall curve (PR-
Curve) without and with proposed feature selection method. Because of these factors, the Naive Bayes
followed by SVM model did the best without the proposed feature selection method (97.05% and 96.47%
accuracy), and the Naive Bayes followed by k-NN model did the best with the proposed feature selection
method on the WBCD (99.41% and 98.82% accuracy) for the breast cancer classification model.