Improving missing-value estimation in microarray data with Collaborative Filtering based on rough-set theory

Data mining techniques have been used to extract useful knowledge from DNA microarray gene expression data for discovering the relations between novel diseases and their related genes. However, DNA microarray gene expression data often contain missing values that must be dealt with to prevent them from signicantly affecting analysis results. Hence, a number of missing-value imputation approaches have been proposed. In this paper, an intelligent imputation approach named the CFBRST (Collaborative Filtering Based on Rough-Set Theory) method is proposed to impute missing values more accurately than currently done by existing approaches. Experimental results on real microarray gene expression datasets reveal that the proposed approach can effectively improve missing-value estimation. The collaborative filtering (CF) approach is often used in recommender systems due to its excellent performance. The proposed CFRBS method is based on the CF method and rough-set theory. The CFBRST method is compared with the k-nearest neighbor (k-NN) imputation algorithm. Experimental results show that the CFBRST method has better accuracy than that of a k-NN approach for yeast cDNA microarray datasets, especially when the percentage of missing values is high.