Research On Feature Selection Based On Heuristic Algorithm

With the advent of the era of big data,people are facing larger and larger volumes of data,but the value density of data is getting smaller and smaller,which makes it more and more difficult to mine value from data.In machine learning,objects usually contain a large number of features,of which only a small part of the features may affect the learning results.Selecting useful features from a large number of features can effectively reduce the data dimension and improve the efficiency of machine learning.However,the number of subsets of the feature set increases exponentially with the increase in the number of features.Traditional feature selection methods cannot cope with the problem of solving spatial combination explosion in a big data environment.In view of the difficulties in high-dimensional feature selection,this paper uses the heuristic algorithm's efficient search capability to propose a feature selection method based on heuristic algorithm,and a heuristic algorithm-based training method for generating adversarial networks The combination is used for imbalanced data classification.The main results and innovations of this article are as follows:(1)An hybrid grey wolf and water wave optimization algorithm is proposed,which Mix the original Grey Wolf Optimization and Water Wave Optimization.Without affecting the ability of the original Grey Wolf algorithm,it enhances its local search capabilities and introduces its application in In feature selection,experimental results show that the method significantly improves the classification accuracy of ordinary classifiers.(2)A Wasserstein generative adversarial network model(WGAN-EFS)based on feature selection is designed.The feature selection of data is added to the adversarial training of the generator and discriminator,and the structure of the adversarial network is optimized by using evolutionary algorithms.Tested a variety of evolutionary algorithms.Among them,the ecogeography-based optimization(EBO)showed better optimization ability to generate the structure of the adversarial network and the feature selection of the data,thereby modeling the model for high-dimensional imbalanced data classification.The proportion of unbalanced data is changed through this network.The experimental results show that this method significantly improves the classification performance of unbalanced data.