A New Unsupervised Feature Ranking Method for Gene Expression Data Based on Consensus Affinity

摘要: Feature selection is widely established as one of the fundamental computational techniques in mining microarray data. Due to lack categorized information practice, unsupervised feature more practically important but correspondingly difficult. Motivated by cluster ensemble techniques, which combine multiple clustering solutions into a consensus solution higher accuracy and stability, recent efforts proposed use these oracles. However, methods are dependent on both particular algorithm used knowledge true number. These will be unsuitable when number not available, common practice. In view above problems, new ranking method evaluate importance features based affinity. Different from previous works, our compares corresponding affinity each between pair instances matrix solutions. As result, alleviates need know clusters dependence approaches works. Experiments real gene expression data sets demonstrate significant improvement results compared several state-of-the-art techniques.