Parallel genetic algorithm of feature selection for complex system analysis

Abstract

The paper shows the approach of important features selecting. These features characterize the evolution of a complex system. Sets of features can include unimportant features. We proposed a parallel genetic algorithm to solve this problem. This algorithm uses short evolutionary paths. It makes algorithm of features selection not so long and in short time separates important features from noisy. As a result of the proposed approach, we carried out search for the best number of parallel evolutionary paths for the feature selection. We demonstrated effectiveness of the proposed approach on the basis of the data analysis of production enterprise functioning. This paper also shows comparison results of parallel genetic algorithm with other algorithms of feature selection by standard deviation, Fisher criterion and multiple determination coefficient. The proposed method for constructing models of complex systems is based on a multidimensional nonlinear regression model using the methods of group accounting of arguments, a parallel genetic algorithm for selecting important features. The proposed method allows to obtain higher quality aggregate input features. Application of parallel genetic algorithm and artificial intelligence methods allows you to generate the regression equation, allowing to make a qualitative forecast of development of complex systems. It could be company or sociological object.