Using background knowledge with attribute-oriented data mining

It is frequently the case that data mining is carried out in an environment which contains noisy and missing data. This a particularly likely to be true when the data were originally collected for a different purpose, as is often the case in data warehousing. The provision of tools to handle such imperfections in data has been identified as a challenging area for knowledge discovery in databases (Fayyad et al., 1996). Previous work has provided some methods of handling such data using machine learning or statistical methods to predict likely values to replace the missing or noisy values. Generalised databases have been proposed to provide intelligent ways of storing and retrieving data. Frequently, data are imprecise, i.e. one is not certain about the specific value of an attribute but only that it takes a value which is a member of a set of possible values. Such data have previously been discussed as a basis of attribute-oriented induction for data mining (Han and Fu, 1996). This approach has been shown to provide a powerful methodology for the extraction of different kinds of patterns from relational databases. It is therefore important that appropriate functionality is provided for database systems to handle such information. The authors consider the problem of aggregation for such data