Fuzzy Rules for Large Feature Spaces

Many relevant applications of information mining e.g. analysis of gene expression
patterns inherently deal with high dimensional data. Similarly multimedia data
analysis and multi-relational database mining usually involve dealing with high
dimensional feature spaces.

In addition to that, data fusion and the propositionalization of multirelational
databases, which are essential to accessing data previously unavailable for many
analysis methods, usually produce high dimensional data sets. Although the
number of input variables can sometimes be reduced with preprocessing, it is
necessary to include methods that are robust to high  dimensional input data.
But while large feature spaces have to be searched, interesting relationships in
such data often involve smaller subsets of  variables and can comprehensibly
be represented by fuzzy rules.

Fuzzy rules are easily understood by the users thus fulfi lling their immediate
information needs. The fuzzy rule induction algorithm given in is speci cally suited
to dealing with large feature spaces and heterogeneous data. An advanced
version constructs a hierarchy of rule sets with different levels of complexity.
For instance this approach allows users to assess basic relations with
relatively coarse fuzzy rules while using a higher level of detail for prediction tasks.