Abstract: Generalization ability is a principal issue in the field of machine learning. Feature weighting is a general case of feature selection, which has the potential of performing better (or at least similar) feature selection. This paper proposes a new ensemble method named FWEART (Feature Weighting based Ensemble of Adaptive Resonance Theory networks), in which an adaptive genetic algorithm is used to conduct a search for the weight vector that can optimize the classification accuracy of the individual Category ART networks. Furthermore, the generalization ability of the ensemble is improved. Experiments on the UCI datasets show that FWEART has good generalization ability. Finally, FWEART is applied to predict the types of earthquake. The result is satisfactory.

Key words: category ART neural network, earthquake type prediction, feature weighting, ensemble learning