Informative attributes selection in nonparametric regression estimation by making use of genetic algorithms

A method of informative attributes selection in nonparametric regression estimation based on genetic algorithms is considered. The idea of the method consists in optimization of attributes fuzzy parameters using genetic algorithms and elimination of attribute with maximum value of fuzzy parameter. Investigation of the method for problems with different dimension (3, 5, 7, and 9), without noise and with 10% noise, for different setting of genetic algorithm parameters was performed. Special attention was paid to investigation of comparative efficiency for different mutation types at genetic algorithm. It is possible to draw following conclusions based on numerical experiments: 1) The method defines the least informative attribute. 2) Noise is not essential for efficiency of the method. 3) Different settings of genetic algorithm parameters for different problems can be effective. So the problem of genetic algorithm parameters setting is actual.

Keywords

Authors

References