Accuracy-based Classifier Systems Using Evolutionary Neural Networks Representation

Accuracy-based classifier systems (XCS) traditionally use a binary string rule representation with wildcards added to allow for generalization over the population encoding. However, the simple scheme has some of drawbacks in complex problems. A neural network-based representation is used to aid their use in complex problem. Here each rule's condition and action are represented by a small network evolved through the action of the genetic algorithm. Also in this work a second neural network is used as classifier's prediction, trained by back propagation. After describing the changes required to the standard XCS functionality, the results are presented using neural network to represent individual rules. Examples of use are given to illustrate the effectiveness of the proposed approached.

Keywords:

Accuracy-based classifier systems , neural networks , rule representation