Feature Grouping And Application Based On Biclustering

With the rapid development of science and technology, a variety of information is filled with people's work and life, hoping to discover useful knowledge from these vast amounts of information. The emergence of data mining technology to provide people with effective knowledge discovery techniques. And has become a very important subject. Data mining techniques are applied in Bioinformatics widely, especially for the biological data obtained by biological analysis often have high dimension features, useful information can be extracted from these data through data mining techniques.Biological data obtained through biological, analysis often exist a large number of redundant or noise features. Within the process of feature selection on biological data, we hope to filter out the noise features and to minimize the presence of redundant features. The use of clustering methods to cluster features, features with high correlations are assigned to the same cluster, then selecting features from these feature clusters is a solution. However, traditional clustering methods in the clustering features only can find the related features in the global feature space. If there are some samples, which may be due to biological analysis techniques or the samples themselves, cause deviation of the data, turn into specific points, and making some of features only have correlations in the majority of the sample space, rather than the global sample space. The use of traditional clustering methods to handle such data can't get a good clustering results. The paper proposed a method that group features by biclustering algorithm, and which can group features self-adaptively, and ensure the features which are grouped into the same group have the similarity in most of the sample space, solved the problems which traditional clustering methods have. After completing the step of grouping features, extracting features from these feature groups for multiple classifiers to establish the corresponding classification models, and forming an ensemble classification method by ensemble strategy.Through the four public microarray data sets and a metabolomics liver data tests, to validate the performance the proposed algorithm, and compared to another ensemble classification method based grouping features, the final experiment results have showed a good classification performance.