Mining Information from Model Build Without Information Disclosure

Abstract

This paper confers an approach of categorizing multiple datasets that are distributed at various sites. The variation in this method of classification is that all the parties jointly build a decision tree model revealing only sufficient information and hiding superfluous data. We have used secure protocols such as secure sum and secure union using commutative encryption technique while model construction. The process discussed in our paper builds efficient, binary classification trees which can in turn predict a test data given by any of the parties involved in the classification.