Performance analysis of combined algorithms for hybridization in mammography

The main purpose of this study is to observe the accuracy improvement of algorithm hybridization and to select which combination among candidate algorithms can provide the best improvement in breast cancer diagnosis. The classifier candidates are Naïve Bayes, Sequential Minimal Optimization, Multilayer Perceptron, C4.5, and Rough Sets. The selection of classifier combination is based on two major factors. The first factor is the maximum accuracy improvement and the second factors are the sensitivity, ROC area under curve, and specificity of each classifier. This study shows that C4.5, Rough Sets, and Naïve Bayes outperform other algorithms in terms of sensitivity, specificity, and ROC area under curve respectively. A combination which comprises Naïve Bayes, Multilayer Perceptron, C4.5, and Rough Sets outperforms other possible combination. By using this combination, there is an improvement of 7.8219% accuracy maximally.