SECURITY IN STATISTICAL DATABASES FOR QUERIES WITH SMALL COUNTS.

Abstract

The security problem of statistical databases containing anonymous but individual records which may be evaluated by queries about sums and averages is considered. A model, more realistic than the previous ones, is proposed, in which nonexisting records for some keys can be allowed. Under the assumption that the system protects the individual's information by the well-known technique which avoids publishing summaries with small counts, several properties about the system and a necessary and sufficiernt condition for compromising the database have been derived. The minimum number of queries needed to compromise the database is also discussed.