Privacy Preserving RFE-SVM for Distributed Gene Selection
The Support Vector Machine Recursive Feature Elimination (SVMRFE) is one of the most effective feature selection methods which has been successfully used in selecting informative genes for cancer classification. This paper extends this well-studied algorithm to the privacy preserving distributed data mining issue. For gene selection over multiple patient data from different sites, the authors propose a novel RFE-SVM method which aims to learn global informative gene subset to get the highest cancer classification accuracy, with limits on sharing of information. They experiment it using Leukemia bio-medical dataset.