The Elimination Of Inference Channel Based On Rough Set Theory

With the proliferation of computer and network technology, database security issues becoming increasingly prominent. So database security has become an important research area of information security. The research of inference problems in databases is a key field of high secure level database system and is an important means to achieve database security.Database inference is a security concern because serious compromises can occur from "inference attacks". An inference attack occurs when a user is able to infer sensitive information from authorized query responses and prevailing common knowledge. However, with database security issues is getting ever more acute, the traditional security measures (such as static inference control mechanism, etc.) have been unable to protect database security. Rough set technology is a new security technology, it data analysis uses only internal knowledge, avoids external parameters, and does not rely on prior model assumptions such as probabilistic distribution in statistical methods, membership function in fuzzy sets theory. Its basic idea is to unravel an optimal set of decision rules from a decision information system.So we use rough set in order to eliminate inference.This dissertation discusses mainly using rough set theory on the basis of reading a lot of references. The paper discusses three problems as follows:Firstly, this paper addresses inference channel problem based on rough set theory.Secondly, a novel rough set approach for discovering inference rules from decision information system is proposed. The approach involves the formulation of a knowledge induction procedure to identify inference rules with a minimal set of features (a reduct) for classification. By using attribute reduct, the classification results coincide with the ones obtained by using all of the attributes. Then we use attribute value reduct, at the same time, we proposed the Value Reduction Matrix. We can obtain the simplest inference rules by this matrix.Thirdly, we discover an algorithm for obtaining inference rules by applying above theorem .First we define the thresholds of Support of the rule and Confidence of the rule Ssup, Sconf, and find out the rules which have higher thresholds than we have defined.The rules we obtain called inference rules. We improve the security level of condition attributes which has high attribute frequency to the security level of decision attributes in inference rules. Thus we can eliminate the inference channel.Finally,the result of our experiment shows that the algorithm of eliminating inference channel is efficient and feasible.