Predictive Adaptive Resonance Theory and Knowledge Discovery in Databases
This paper investigates the scalability of predictive Adaptive Resonance Theory (ART) networks for knowledge discovery in very large databases. Although predictive ART performs fast and incremental learning, the number of recognition categories or rules that it creates during learning may become substantially large and cause the learning speed to slow down. To tackle this problem, we introduce an on-line algorithm for evaluating and pruning categories during learning. Benchmark experiments on a large scale data set show that on-line pruning has been effective in reducing the number of the recognition categories and the time for convergence. Interestingly, the pruned networks also produce better predictive performance.
Unable to display preview. Download preview PDF.
- 3.Kohavi, R.: Scaling up the accuracy of naive-bayes classifiers: a decision-tree hybrid. Proceedings, KDD-96 (1996)Google Scholar
- 4.Murphy, P. M., Aha, D. W.: UCI repository of machine learning databases [machine-readable data repository]. Irvine, CA: University of California, Department of Information and Computer Science (1992)Google Scholar