Feature region-merging based fuzzy rules extraction for pattern classification
Hongwei Zhu, Otman Basir · 2004
A supervised learning method is proposed to automatically extract fuzzy rules for numerical pattern classification problems. fuzzy rules are constructed corresponding to hyperboxes in a multi-dimensional feature space, where a hyperbox indicates an existence region of data belonging to a singleton class or a compound class. Hyperboxes are effectively realized by means of a linked list based region-merging technique. The method supports the representation of the union of multiple classes in the region merging process and hence it can deal with compound classes in the cases where highly mixed classes exist. Also, the method is capable of automatically deleting trivial features during the rule learning process. To demonstrate the effectiveness of the proposed method, experiments are carried out for classifying Iris data set and human brain magnetic resonance images (MRI). It is concluded that the proposed method performs well and is quite competitive to other fuzzy rule extraction techniques.