Prediction of Breast Cancer Survivability using Ensemble Algorithms

In this paper we propose new ensemble cancer survivability prediction models based three variants of AdaBoost algorithm to extend the application range of ensemble methods. In our approach to address the problem of low efficiency and slow speed we use Random Forest, Radial Basis Function and Neural Network algorithms as base learners and AdaBoostM1, Real AdaBoost and MultiBoostAB as ensemble techniques. AdaBoost is a technique that iteratively trains its base classifiers to generate committee of strong classifiers to improve their performance and prediction accuracy. There has been major research in ensemble modeling in statistics, medicine, technology and artificial intelligence in the last three decades. This might be because of the effectiveness and reliability of the technique in helping medical and other professionals in diagnosis and incident predictions. However, there is a need to improve the accuracy of the existing models address current challenges. In this paper we use state of the art Wisconsin breast cancer dataset in training and testing the proposed hybrid models. The performance of the models was evaluated using the following performance metrics: Accuracy, RMSE, TP Rate, FP Rate, Precision and ROC Area. The results of our study shows that MBAB-RF and AdaM1-RF models have the same accuracy prediction of 97% and RA + ANN has the worst prediction accuracy of 88%. Additionally we found that all ANN models requires more time to train its committee of classifiers compared to RFB models that requires the least time despite the fact that RBF is a family of ANN algorithm.