Evolving the input space for a decision system

This paper addresses the problem of classifying data sets. The aim of this research is to extend the discrimination of a decision system by adding polynomials of the base inputs to the available inputs. The polynomials used to extend the inputs are evolved using the quality of the decision trees resulting from the extended inputs as a fitness function. The system has been integrated into the WEKA framework and so allows other components to be used in the decision system. So far our approach has focused on decision trees using the base inputs and compares it with a decision tree built using the extended input space. NSGA-II, a multi-objective optimisation algorithm, has also been integrated into the system to investigate the effects of added selection pressure on the generated polynomials. Because we have a library of procedures which we can apply we have also used Reduct systems in concert with the extension system and experiments show that although these reduce the input space the overall decision system is no better and can take longer to develop.