Cascade generalization for breast cancer detection

Abstract

Mammography is known as the preferred method for breast cancer diagnosis. Researchers have proposed machine learning based methods to improve the detection of breast cancer using mammography. In this study, cascade generalization is proposed for breast cancer detection. Four Bayesian Network based methods, SVM, and C4.5 are evaluated in loose coupled cascade classifier. The Bayesian based methods are evaluated in both base level and meta level. The evaluation results show the superiority of the proposed cascade strategy compared to Bagging and single classifier approach. Naive Bayes with SMO cascade demonstrated the best result in terms of ROC area under curve of 0.903. Bayesian Network using Tabu search with SMO cascade demonstrated the best accuracy of 83.689%.