Combined mRMR filter and sparse Bayesian classifier for analysis of gene expression data

摘要

Many disorders can be diagnosed by analysis of gene expression microarrays and this can save lots of lives. However, as gene expression data have high dimensions, establishing a method to identify the genes related to the target disease still remains a challenge, because it should provide a well-grounded prediction about the disease status. To this end, the best subset of genes should be distinguished for the classification task. In this paper, we have introduced a new framework for the analysis of gene expression data. Our proposed algorithm tries to find the best feature subset, in two main stages. First, an information theoretic forward feature selection algorithm called mRMR (minimum redundancy, maximum-relevancy) is used to find a candidate set for best features. In the next stage, the RVM (Relevance Vector Machine) classifier which is well suited for gene data analysis is utilized. The RVM has frequent privileges over other classifiers, namely, it can return a membership probability for each class that can be very vital for diagnosis of dramatic diseases, and it can also lead to a more sparse approach to fit a model over the training data which will help to avoid overfitting, etc. The Experimental results showed that the proposed algorithm outperforms the previous works in both classification accuracy and sparsity of the model.