Evolving rules from neural networks trained on continuous data

Abstract

Artificial neural networks (ANNs) are used extensively involving continuous data. However, their application in many domains is hampered because it is not clear how they partition continuous data for classification. The extraction of rules, therefore, from ANNs trained on continuous data is of great importance. The system described in this paper uses a genetic algorithm to generate input patterns which are presented to the network, and the output from the ANN is then used to calculate the fitness function for the algorithm. These patterns can contain null characters which represent a zero input to the ANN, and this allows the genetic algorithm to find patterns which can be converted into additive rules with few antecedent clauses. These antecedents provide information as to where and how the neural network has partitioned the continuous data and can be combined together to make rules. These rules compare favourably with the results of those generated by See5 (a decision tree-based data mining tool) when executed on a data set consisting of continuous attributes.