Evaluating Diagnostic Performance of Machine Learning Algorithms on Breast Cancer

摘要

This paper focuses on comparing performance of six data mining methods namely: Bagging, SVM (SMO), Decorate, C4.5 (J48), Naive Bayes and IBK in analyzing Wisconsin Breast Cancer (WBC) datasets. The datasets were obtained from the UCI Machine Learning Repository and comprises of 699 instances and 11 attributes. A confusion matrix, based on a 10-fold cross validation technique was used in our experiment to provide the basis for measuring the accuracy of each algorithm. We introduce an idea of combining the algorithms at classification level to obtain the most ideal multi-classifier approach for the WBC data set. Waikato Environment Knowledge Explorer (WEKA), open source data mining software was used for the experimental analysis. The experimental results show that SMO offers the best accuracy (97 %) among the six algorithms, while merging SMO, Naive Bayes, J48 and IBK offers the best accuracy (97.3%) on the data set.