Efficient GA based techniques for classification

A common approach to evaluating competing models in a classification contex t is via accuracy on a test set or on cross-validation sets. However, this can be computationally costly when using genetic algorithms with large data sets and the benefits of performing a wide search are compromised by the fa ct that estimates of the generalization abilities of competing models are s ubject to noise. This paper shows that clear advantages can be gained by us ing samples of the test set when evaluating competing models. Further, that applying statistical tests in combination with Occam's razor produces pars imonious models, matches the level of evaluation to the state of the search and retains the speed advantages of test set sampling.