Analysis of Breast Cancer Detection Using Different Machine Learning Techniques

Abstract

Data mining algorithms play an important role in the prediction of early-stage breast cancer. In this paper, we propose an approach that improves the accuracy and enhances the performance of three different classifiers: Decision Tree (J48), Naïve Bayes (NB), and Sequential Minimal Optimization (SMO). We also validate and compare the classifiers on two benchmark datasets: Wisconsin Breast Cancer (WBC) and Breast Cancer dataset. Data with imbalanced classes are a big problem in the classification phase since the probability of instances belonging to the majority class is significantly high, the algorithms are much more likely to classify new observations to the majority class. We address such problem in this work. We use the data level approach which consists of resampling the data in order to mitigate the effect caused by class imbalance. For evaluation, 10 fold cross-validation is performed. The efficiency of each classifier is assessed in terms of true positive, false positive, Roc curve, standard deviation (Std), and accuracy (AC). Experiments show that using a resample filter enhances the classifier’s performance where SMO outperforms others in the WBC dataset and J48 is superior to others in the Breast Cancer dataset.