Data Confidentiality and Chase-Based Knowledge Discovery

This article discusses data security in Knowledge Discovery Systems (KDS). In particular, we presents the problem of confidential data reconstruction by Chase (Dardzinska and Ras, 2003c) in KDS, and discuss protection methods. In conventional database systems, data confidentiality is achieved by hiding sensitive data from unauthorized users (e.g. Data encryption or Access Control). However, hiding is not sufficient in KDS due to Chase. Chase is a generalized null value imputation algorithm that is designed to predict null or missing values, and has many application areas. For example, we can use Chase in a medical decision support system to handle difficult medical situations (e.g. dangerous invasive medical test for the patients who cannot take it). The results derived from the decision support system can help doctors diagnose and treat patients. The data approximated by Chase is particularly reliable because they reflect the actual characteristics of the data set in the information system. Chase, however, can create data security problems if an information system contains confidential data (Im and Ras, 2005) (Im, 2006). Suppose that an attribute in an information system S contains medical information about patients; some portions of the data are not confidential while others have to be confidential. In this case, part or all of the confidential data in the attribute can be revealed by Chase using knowledge extracted at S. In other words, self-generated rules extracted from non-confidential portions of data can be used to find secret data. Knowledge is often extracted from remote sites in a Distributed Knowledge Discovery System (DKDS) (Ras, 1994). The key concept of DKDS is to generate global knowledge through knowledge sharing. Each site in DKDS develops knowledge independently, and they are used jointly to produce global knowledge without complex data integrations. Assume that two sites S1 and S2 in a DKDS accept the same ontology of their attributes, and they share their knowledge in order to obtain global knowledge, and an attribute of a site S1 in a DKDS is confidential. The confidential data in S1 can be hidden by replacing them with null values. However, users at S1 may treat them as missing data and reconstruct them with Chase using the knowledge extracted from S2. A distributed medical information system is an example that an attribute is confidential for one information system while the same attribute may not be considered as secret information in another site. These examples show that hiding confidential data from an information system does not guarantee data confidentiality due to Chase, and methods that would protect against these problems are essential to build a security-aware KDS.