GFAM: A Genetic Algorithm Optimization of Fuzzy ARTMAP A. Al-Daraiseh, M. Georgiopoulos, G. Anagnostopoulos, A. S. Wu, M. Mollaghasemi Abstract— Fuzzy ARTMAP (FAM) is currently considered to be one of the premier neural network architectures in solving classification problems. One of the limitations of Fuzzy ARTMAP that has been extensively reported in the literature is the category proliferation problem. That is Fuzzy ARTMAP has the tendency of increasing its network size, as it is confronted with more and more data, especially if the data is of noisy and/or overlapping nature. To remedy this problem a number of researchers have designed modifications to the training phase of Fuzzy ARTMAP that had the beneficial effect of reducing this phenomenon. In this paper we propose a new approach to handle the category proliferation problem in Fuzzy ARTMAP by evolving trained FAM architectures. We refer to the resulting FAM architectures as GFAM. We demonstrate through extensive experimentation that an evolved FAM (GFAM) exhibits good generalization, small size, and produces an optimal or a good sub-optimal network with a reasonable computational effort. Furthermore, comparisons of the GFAM with other approaches, proposed in the literature, that address the FAM category proliferation problem, illustrate that the GFAM has a number of advantages (i.e. produces smaller or equal size architectures, of better or as good generalization, with reduced computational complexity). I.INTRODUCTION HE Adaptive Resonance Theory (ART) was developed by Grossberg (1976). One of the most celebrated ART architectures is Fuzzy ARTMAP (Carpenter et al, 1992), which has been successfully used in the literature for solving a variety of classification problems. One of the limitations of Fuzzy ARTMAP (FAM) that has been repeatedly reported in the literature is the category proliferation problem, which is tightly connected with the issue of overtraining. Manuscript received January 31, 2006. This work was supported in part by the National Science Foundation (NSF) under grants CRCD 0203446 and CCLI 0341601. A. Al-Daraiseh is with the School of Electrical Engineering and Computer Science, University of Central Florida, Orlando, FL 32816, USA (e-mail: creepymaster@yahoo.com ). M. Georgiopoulos is with the School of Electrical Engineering and Computer Science, Orlando, FL 32816, USA (phone: (407) 823-5338, fax: (407) 823 5835; e-mail: michaelg@mail.ucf.edu ). G. Anagnostopoulos is with the Department of Electrical and Computer Engineering, Florida Institute of Technology, Melbourne, FL 32901, USA (e-mail: georgio@fit.edu ). A. S. Wu is with the School of Electrical Engineering and Computer Science, University of Central Florida, Orlando, FL 32816, USA (e-mail: aswu@cs.ucf.edu ). M. Mollaghasemi is with the Department of Industrial Engineering and Management Systems, University of Central Florida, Orlando, FL 32816, USA (e-mail: mollagha@mail.ucf.edu ). A number of authors have tried to address the category proliferation/overtraining problem in Fuzzy ARTMAP. Amongst them we refer to the work by Verzi, et al., 2001, Anagnostopoulos, et al., 2003 and Gomez-Sanchez, et al., 2001, where different methods were introduced and evaluated, that allow Fuzzy ARTMAP categories to encode patterns that are not necessarily mapped to the same label. In this paper, we propose the use of genetic algorithms (Goldberg, 1989) to solve the category proliferation problem in Fuzzy ARTMAP. Genetic algorithms (GAs) are a class of population-based stochastic search algorithms that are developed from ideas and principles of natural evolution. An important feature of these algorithms is their population based search strategy. Individuals in a population compete, modify and exchange information with each other in order to perform certain tasks. Our approach starts with a population of trained FAMs. GA operators are then utilized to manipulate these trained FAM architectures in a way that encourages better generalization and smaller size architectures. The evolution of trained FAM architectures allows these architectures to exchange and modify their categories in a way that emphasizes smaller and more accurate FAM architectures. Eventually, this process leads us to a FAM architecture (referred to as GFAM) that has good generalization performance and creates networks of small size; all of these benefits come with the additional advantage of reasonable computational complexity. Genetic algorithms have been extensively used to evolve artificial neural networks. For a thorough exposition of the available research literature in evolving neural networks the interested reader is advised to consult Yao, 1999. To the best of our knowledge there is no work conducted in the literature so far that has attempted to evolve FAM neural network structures, and that is the main focus of our effort. The organization of this paper is as follows: In section 2 we present GFAM. In Section 3, we describe the experiments and the datasets used to assess the performance of GFAM, and we also compare GFAM to four other ART networks that attempted to resolve the category proliferation problem in Fuzzy ARTMAP. Finally, in Section 4, we summarize our work. T 0-7803-9489-5/06/$20.00/©2006 IEEE 2006 IEEE International Conference on Fuzzy Systems Sheraton Vancouver Wall Centre Hotel, Vancouver, BC, Canada July 16-21, 2006 315