Tribal evolution competition-based characteristic selection method

Abstract

The invention discloses a tribal evolution competition-based characteristic selection method, which is used for solving the technical problem of offset selected characteristic number in permutation and combination during high-dimensional characteristic classification in an existing characteristic selection method. According to the technical scheme, the method comprises the steps of dividing characteristics into a plurality of tribes according to characteristic dimensions, wherein the number of the characteristics selected in the tribes follows the Gaussian distribution in statistics; ensuring a distribution situation of the number of the characteristics selected in the tribes not to be changed on the premise of ensuring an optimal individual to be reserved by adopting an improved genetic algorithm; and enabling the tribes to start competition after each tribe evolves to a certain extent, providing an elite individual with the highest fitness in the tribe by each tribe, and sorting the elite individuals of all the tribes according to the fitness. The characteristic space is divided into the Gaussian tribes, the number of the characteristics selected in the tribes follows the Gaussian distribution in statistics, and the Gaussian distribution is always kept unchanged, so that the offset problem can be effectively solved and a global optimal subset can be searched for.

Description

technical field [0001] The invention relates to a feature selection method, in particular to a feature selection method based on tribal evolution competition. Background technique [0002] The document "Research of multi-population agent genetic algorithm for feature selection, Expert Systems with Applications, 2009, Vol36(9), p11570-11581" proposes a feature selection algorithm based on multi-population genetic evolution. The number of original features is N. This method divides the feature space into multiple connected regular polygons by constructing a polygonal closed-loop population, and selects individuals whose number of features is M (M<N) through a competition mechanism, so that the classification effect is the best. The population communicates through two shared individuals. These two individuals are called population agents. The agents and other members of the population are sorted according to their fitness. Those with high fitness are retained and those with ...

AI Technical Summary

Problems solved by technology

When the method described in the literature constructs multiple populations, it does not take into account the bias of the number of feature selections in the permutation and combination. The dimension of the selected feature subset is likely to be around N / 2, because there are M features The number of possible feature subsets composed of When N is large, the number of features selected by individuals in the population has a very high probability of falling near N / 2, making the final result likely not to be globally optimal

Images