Breast Cancer Classification using Deep Learning-based Ensemble

Abstract

Objectives: We propose a deep learning-based ensemble for improving breast cancer classification and compare it with existing six models including deep neural network on two UCI data. Methods: We propose a deep learning-based stacking ensemble method. We first applied five classifications methods individually, which were k-nearest neighbor, decision trees, support vector machines, discriminant analysis, and logistic regression analysis and then adopt a deep learning to the predictions derived from these methods after using 5-fold cross validation technique. We compared the proposed deep learning-based ensemble method with these methods for two UCI data through classification accuracy, ROC curves and c-statistics. Results: Experimental results for two UCI data showed that the proposed deep learning-based ensemble outperformed single k-nearest neighbor, decision trees, support vector machines discriminant analysis, and logistic regression analysis as well as deep neural network in terms of various performance measures. Conclusions: We proposed deep learning-based ensemble for improving breast cancer classification. The deep learning-based ensemble outperformed existing single models for all applications in terms of various performance measures.