Research And Application Of Feature Selection Method Based On Fruit Fly Optimization Algorithm

Feature selection is a popular method to reduce the number of high-dimensional data attributes,and it is an essential pre-processing work in classification.With the development of computer and data storage technology,data volume increases significantly.Ideally,the information provided is valid,but in practice,the data often contains irrelevant,redundant,and noisy information.The presence of these features in the dataset may mislead learning algorithms or lead to excessive data matching,and feature selection can help data mining and machine learning algorithms work faster and more efficiently.The meta-heuristic algorithm is a general heuristic algorithm.Under the condition of limited time and cost,the meta-heuristic algorithm is helpful to find the approximate optimal solution from the vast solution space in limited time.Recently,nature-inspired meta-heuristic algorithms have received a lot of attention for solving feature selection problems because they can use good exploration and development strategies to find competitive solutions.The representative methods of natural heuristic feature selection to find the optimal subset are grey Wolf optimization algorithm,firefly optimization algorithm,particle swarm optimization algorithm,whale optimization algorithm,etc.In this paper,the fruit fly optimization algorithm is used as the search strategy,and a package feature selection method is designed.In order to alleviate the problem that the original fruit fly optimization algorithm is prone to premature convergence and local optimization when dealing with the feature selection problem,we propose four different variants of fruit fly optimization algorithm based on the concept of evolutionary population dynamics,and at the same time,we put forward the mutation operator and crossover operator based on the group dynamics in evolution.The algorithm was tested on 20 benchmark data sets.The comparison results show that the drosophila fly optimization algorithm combined with the competition selection operator has a significant influence on the feature selection of the original binary drosophila fly,which can enhance the optimization ability of the original algorithm,make it better than other optimization algorithms,and find the best choice.Through the analysis of BIFFOA_EPD_Tour,we found that the algorithm has limitationsin the initialization and search stages.Improvement of the above two aspects may improve the algorithm's searchability and dimension reduction ability.Therefore,we improved the BIFFOA_EPD_Tour and proposed a fly optimization algorithm combining mutual information(MIFFOA for short).We used the initialization strategy based on mutual information to delete some features with poor performance,reduce the dimension and reduce the computation of the subsequent algorithm while ensuring the classification accuracy.At the same time,a local elite search strategy based on mutual information is proposed to strengthen the development ability of the algorithm.In the first stage,more information features are added to improve the classification accuracy of the classifier.The second phase is to reduce the number of features selected while maintaining high classification accuracy.The experimental results show that compared with BIFFOA_EPD_Tour,MIFFOA can achieve higher classification accuracy and stronger dimension reduction ability without reducing classification accuracy.