Minimization of side effects in sensitive information hiding

Abstract

The improvement of data storing technologies and the spread of the internet made the variety of databases to be everywhere. To find meaningful information in databases of several area, data mining methodology also quickly advanced. However, through data mining technique user’s private data or confidential information of company can be leaked. Get over this difficulty, Privacy preserving data mining (PPDM) came to the fore. The objective of PPDM is to prevent sensitive information leaks while minimize the loss of database. In this thesis, solving the problem of ‘frequent itemset hiding’, one of basic privacy preserving challenges, the framework is proposed. To sanitize the database efficiently, Integer programming model is introduced. The results show that impressive constraint reduction are achieved by proposed method. In excrement, roughly 30~40% of constraints in the whole constraints sets are reduced and most of trials in all experiments have one iteration, which mean infeasible constraints are removed with minimized loss.