On discovering functional relationships when observations are noisy: 2D Case
A heuristic method of model selection for a nonlinear regression problem on R 2 is proposed and discussed. The method is based on combining nonparametric statistical techniques for generalized additive models with an implementation of the Equation Finder of Zembowicz and Żytkow (1992). Given the inherent instability of such approaches to model selection when data are noisy, a special procedure for stabilization of the selection is an important target of the method proposed.
Unable to display preview. Download preview PDF.
- Breiman, L.: Bagging predictors. Machine Learning 26 (1996b) 123–140Google Scholar
- Ćwik, J. and Koronacki, J.: A heuristic method of model choice for nonlinear regression. In: Rough Sets and Current Trends in Computing, L. Polkowski and A Skowron (eds.), Springer, LNAI 1424 (1998) 68–74Google Scholar
- Krzanowski, W.J.: Principles of multivariate analysis: A user’s perspective. Oxford Univ. Press, 1988.Google Scholar
- Moulet, M.: Comparison of three inductive numerical law discovery systems. In: Machine learning and statistics, G. Nakhaeizadeh and C.C. Taylor (eds.), Wiley, 1997, 293–317Google Scholar
- Zembowicz, R. and Żytkow, J.M.: Discovery of equations: Experimental evaluation of convergence. In: Proceedings of the 10th National Conference on Artificial Intelligence, AAAI Press, 1992, 70–75Google Scholar