| Breast cancer is one of the most prevalent tumor diseases among women, which has received increasing attention in the society. As the breast cancer data is of high dimensions and diversity characteristics, the performance of traditional classifier often failed, i.e. the disaster the dimensions problems. Ensemble learning algorithm is one of the most important and active research directions in pattern recognition and machine learning. It is actually a systematic approach which constructs a integration classifier by multiple classifiers and uses the integration classifier to discriminate the unknown samples.In this paper, a novel approach based on the Dynamic Feature Subset Selection and the EM algorithm with Naive Bayesian classifier integration algorithm (DSFS+EMNB) is proposed to use the rich information of the breast data effectively and also to alleviate curse of dimensionality. The algorithm has to use RSM (Random Subspace Method), Bagging methods, integrated voting method, feature subset selection method (Filter method and Wrapper methods), etc. to construct a dynamic feature subset integrated classifier. And the EM-based Bayesian classifier is employed as the base classifier. for construct ion of the dynamic integration algorithm.The experimental results demonstrated that this method significantly outperforms the other methods like SVM and other traditional methods (EM-Naive Bayes (EMNB), KNN, Boost C5, Neural Net, etc.) in terms of average accuracy, as well as generality. Furthermore, the algorithm has better performance and generalization ability, with the value of further research. The experimental results show that the algorithm has very better performance than traditional classifiers, and the experimental results verify that ICA(Independent Component Analysis) helps improve DSFS+EMNB classification performance for mammographic data, This algorithm can improve classification accuracy and dimension reduction has a certain role. |
PDF Full Text Request