Ensembles of diverse classifiers using synthetic training data

The goal of an ensemble construction with several classifiers is to achieve better generalization than that of a single classifier. And proper diversity among classifiers is considered as the condition for an ensemble construction. This paper investigates synthetic pattern for diversity among classifiers. It alters input feature values of some patterns with the values of other patterns to get synthetic patterns. The pattern generation from using exiting patterns seems emphasize both accuracy and diversity among individual classifiers. Ensemble based on the synthetic patterns is evaluated for both neural networks and decision trees on a set of benchmark problems and was found to show good generalization ability.