A novel feature decomposition method to develop multi-hierarchy model

The comprehensibility of a model is very important since the results should be ultimately be interpreted by a human. This paper presents a new machine learning method, named feature decomposition method based on rough set theory, to discover concept hierarchies and develop a multi-hierarchy model of database. First the features with more relations are selected into a feature group. Then some measures by rough set theory are presented in this paper. According to these measures, the objects defined on the proposed feature group are labeled to discover a new concept. The new concept hierarchies of the database usually have specific meaning, which increase the transparency of data mining process. Finally the rule induction can process on the concept hierarchies of the database to develop a new multi-hierarchy model. The idea presented is illustrated with examples and datasets from UCI machine learning repository. The results show that the multi-hierarchy model established by feature decomposition method can get high classification accuracy and have better comprehensibility.