Using Genetic Algorithm for Breast Cancer Feature Selection

Abstract

Breast cancer has been one of the most widespread cancer types in women worldwide. Breast cancer can be treated when detected early; otherwise, it has one of the highest mortality rates among cancer types. Many tools can be used for detection, but computer-based diagnosis systems have become popular as they are cheaper and quicker. This brings incorrect detections as well. Hence, feature selection is an important factor that can enhance the accuracy of computer-based programs. This study uses genetic algorithms for feature selection within a wrapper methodology for breast cancer diagnosis. The proposed model has been tested with 17 different classifiers in order to evaluate its effectiveness. There has been an increase in training accuracy after feature selection was employed with genetic algorithms. The highest training accuracy was reported in Extra Trees, MLP, Random Forest, and Logistic at 100%, and the lowest was reported in GaussianNB at 0.925. Furthermore, feature selection improved validation accuracy, sensitivity, specificity, F1-score, Matthews Correlation Coefficient, specificity, and sensitivity.