Multilayer Fuzzy ARTMAP
Tatt Hee Oong, Nor Ashidi Mat Isa · 2012
This paper presents a new neural network architecture called the Multilayer Fuzzy ARTMAP (ML-FAM) together with the neighborhood learning algorithm and N-best rule for fast learning and testing in solving pattern classification problem. An analysis to the Fuzzy ARTMAP learning algorithm is studied to identify the weakness of it. ML-FAM uses layered structure to seek for important region of the category in both learning and testing phase. Thus, its learning time and testing time are reduced. Besides that, N-best rule is introduced to improve the generalization performance of ML-FAM particularly for high dimensional problem by taking the advantage of layered structure of ML-FAM. Experimental results show that ML-FAM is superior to Fuzzy ARTMAP in terms of learning time and testing time while preserving the generalization capability.