Acquiring classification rules by using adaptive resonance theory

We propose two on-line classification methods, ARTMAP_ED and ARTMAP_AW, which are based on adaptive resonance theory. ARTMAP_ED, classifies cases on the basis of Euclidean distance and it incorporates category merging as a generalization technique. ARTMAP_AW is the modification of ARTMAP_ED to consider the importance of each attribute. The importance of attributes is updated through generalizing and specializing classification rules. Experimental results show that ARTMAP_AW acquires better classification rules with fewer categories than ARTMAP_ED, fuzzy ARTMAP and C4.5.