Optimization of Non-fuzzy Neural Networks Based on Crisp Rules in Scatter Partition

Abstract

We introduce a design of non-fuzzy neural networks that have crisp rules in scatter partition. To generate the crisp rules and construct the networks, we use hard c-means clustering algorithm. The partitioned local spaces indicate the crisp rules of the proposed networks. The consequence part of the rule is represented by polynomial functions. The coefficients of the polynomial functions are learned using back-propagation algorithm. In order to optimize the parameters of the proposed networks we use particle swarm optimization techniques. The proposed networks are evaluated with the example for nonlinear process.