Abstract
The aim of this research is to analyze nominal data sets. For nominal data sets, it is effective to construct clusters or classifiers based on patterns (conjunctions of attribute values) from the point of view of readability. However, it is important to deal with all of the patterns in a data set, because the number of the patters increases exponentially in proportion to the size of the data. Thus, the authors have proposed a method to deal with the pattern space based on kernel methods and Boolean functions. The proposed kernel is called restricted downward function kernel. In the feature space of the kernel, inconsistent patterns are removed by a Boolean function. To relax the restriction, we consider a relaxed kernel matrix for each object, and add them into one matrix.
