FEATURE SELECTION FOR TIME-SERIES PREDICTION IN CASE OF NONDETERMINED ESTIMATION

Abstract

two sets of features can be compared just with some probability, so the existing methods should be modified. For this purpose we propose the use of Compact Genetic Algorithms (CGA) and present a scheme of feature selection. The step of genetic algorithm learning is modified for the case of stochastic estimation of a feature set. The results for Internet-traffic forecasting are obtained