In this paper, we propose a feed-forward ensemble neural network for data sets having both discrete and continuous attributes. The ensemble provides results that are more accurate than those of conventional neural networks and expresses more comprehensible rules. Through the separation of data in compliance with primary rules, it enables the generation of secondary rules that apply solely to instances of non-compliance with the primary rules and maintain higher accuracy than is conventionally attainable. We demonstrate the high performance of the ensemble neural network with rules extracted by Re-RX, and verify that it can reduce the complexity of handling multiple neural networks.
展开▼
