A hybrid model of fuzzy artmap and the genetic algorithm for data classification

Abstract

A framework for optimizing Fuzzy ARTMAP (FAM) neural networks using Genetic Algorithms (GAs) is proposed in this paper. A number of variables were identified for optimization, which include the presentation order of training data during the learning step, the feature subset selection of the training data, and the internal parameters of the FAM such as baseline vigilance and match tracking. A single configuration of all three variables were encoded as a chromosome string and evaluated by creating and training the FAM according to the variables. The fitness of the chromosome is determined by the final classification accuracy of the FAM. Evaluation on benchmark data sets are conducted with the results compared with literature. Experimental results indicate the effectiveness of the proposed framework in undertaking data classification tasks.