Research On Application Of Hybrid Natural Computing Algorithm In Feature Selection

With the advent of the era of big data,the amount and dimensions of data that people can obtain are also getting higher and higher,the original data must be dimension-reduced during the processing and analysis of the data.The feature selection is a complex combinatorial optimization problem.Therefore,solving the feature selection problem based on natural computing related algorithms has received extensive research and attention.However,theory and practice have proved that no single algorithm can solve all optimization problems.At the same time,the performance of a single algorithm has limitations.To obtain more satisfactory results,two or more algorithms can be used in accordance with certain rules.Forming a hybrid algorithm,while retaining the global or local search characteristics of a single algorithm,it makes up for the relatively weak local or global search capabilities of a single algorithm,and has good application potential in solving complex optimization problems.In this thesis,three algorithms are selected through experimental analysis and comparison,and three hybrid algorithms based on different hybrid rules of the selected algorithm are proposed.At the same time,the hybrid algorithm is applied to the feature selection problem.The main work is as follows:1.Analyzed the limitations of single algorithm independent optimization.By performing unimodal,multimodal and compound function tests on the algorithms involved in the discussion,it is concluded from the overall analysis of the experimental results that it is difficult for a single algorithm to have an excellent overall performance during independent optimization.Search and local development capabilities,and a single algorithm can not adapt to all optimization problems,verifying the necessity and theoretical basis of algorithm mixing.On the other hand,by comparing and analyzing the results of different problems,this thesis summarizes the global search and local development performance characteristics of common natural computing algorithms of different kinds.2.Three hybrid algorithms based on low-level and high-level hybrid ideas are designed.Based on the performance characteristics of common natural computing algorithms summarized,three different hybrid algorithms are designed using differenthybrid ideas.Experiments show that the hybrid algorithm fully embodies the advantages of mixing and collaboration.3.A feature selection method based on hybrid algorithm is proposed,and related applications are made in the classification of public data sets and classification of features in hyperspectral remote sensing images.Hyperspectral remote sensing images need to select bands because of the large amount of data and data bands.The noise bands included in the band set and the excessive number of bands will affect the classification accuracy and may also bring "dimensional disasters".Therefore,applying the feature selection method proposed in this thesis to the band selection problem can effectively test the ability of the proposed method to reduce the data dimension and improve the classification accuracy.In general,the selection of component algorithms in hybrid algorithms,the design of hybrid rules,and the application of hybrid algorithms in feature selection are studied in this thesis.The corresponding experimental results prove that the hybrid algorithm improves the performance of the algorithm and in feature selection.The problem has good performance,high research value and broad application prospects.