A comparison of heuristic methods for polynomial regression model induction

Abstract

We compare four different heuristic methods for polynomial regression model induction. The methods are very different in their approaches. Our main concern in this study is in the differences of candidate model spaces the methods deal with (completely predefined versus non-predefined), as well as search strategies used. We investigate the advantages and disadvantages of the approaches represented by the methods in terms of predictive error, complexity of the induced models and required computational resources. For empirical comparisons, we use twelve test problems.