DISCOVERING MISSING AND UNDERSTANDABLE PATTERNS IN DATABASES

Abstract

While typical discovery techniques in data mining are focused on discovering rules or patterns existing in the data, the discovery of missing patterns can be useful too. Missing patterns are value combinations that occur in no or few data cases. They may not be known before or are unexpected. Knowing those unexpected missing patterns may lead to significant discoveries. Another problem in data mining is that users often find some discovered rules which are not closely related to or are contradicting with their existing concepts, and hence difficult to understand and to accept. Fortunately, the data cases classified under these less understandable rules could often be re-classified under other more understandable but undiscovered ones. The discovery of such understandable rules is very important for practical applications. In this thesis, we propose techniques that aim to ( 1) discover missing patterns or holes in databases, and to (2) transform less understandable classification rules into more understandable ones.