In this paper, we present a methodology that facilitates attribute selection and dependence modelling utilising the Gamma Test and a Genetic Algorithm (GA). The Gamma Test, a non-linear analysis algorithm, forms the basis of an objective function that is utilised within a GA. The GA is applied to a range of multivariate dataseis, describing such problems as house price prediction and lymphography classification, in order to select a useful subset of features (or mask). Local Linear Regression is used to contrast the accuracy of models constructed using full masks and suggested masks. Results obtained demonstrate how the presented methodology assists the production of superior models.
|Enw||Applications and Innovations in Intelligent Systems XI|