Research On Feature Selection Method Based On Improved Self-tuning Adaptive Genetic Algorithm

In the era of big data,feature selection,as a data dimensionality reduction technology,aims to select the most favorable relevant features for the algorithm from the original data,reduce the dimension of data and the difficulty of learning tasks,and improve the efficiency of learning model.Genetic algorithm belongs to random search algorithm,which is not easy to produce local optimal solution.It is also widely used in feature selection research.However,there are still some problems in the research of feature selection: firstly,most of the existing feature selection methods based on genetic algorithm and k-nearest neighbor do not consider the different importance of each feature,and the premature convergence of the feature selection method of traditional adaptive genetic algorithm,especially the problem of local optimal solution;Secondly,the correlation between features is ignored.The final feature subset may not contain some features,resulting in some redundant information in the convergent feature subset.Aiming at the different importance of each feature in most existing feature selection methods based on genetic algorithm and k-nearest neighbor,and the premature convergence of the feature selection method of traditional adaptive genetic algorithm,especially the problem of local optimal solution,this paper proposes a feature selection method based on self-tuning adaptive genetic algorithm and weighted k-nearest neighbor,abbreviated as WKNNSTA?GAFS.This method designs a genetic algorithm which can adaptively adjust the mutation rate,population size and convergence threshold with the population fitness.At the same time,the feature weight vector is used.When calculating the fitness,the weighted k-nearest neighbor algorithm is introduced to calculate the sample category,and each feature is assigned a weight to measure the classification ability of the feature.The self-tuning adaptive genetic algorithm searches the optimal feature weight vector in the iterative evolution process,the best ranking sequence of all features is obtained according to the optimal feature weight vector.In order to evaluate the effectiveness of this method,it is compared with seven existing feature selection methods on five standard data sets.The experimental results show that this method is effective and has high classification performance.Secondly,when WKNNSTA?GAFS method is used for feature selection,the first n features are selected through descending sorting to form the final feature subset.Although the algorithm can effectively eliminate irrelevant features,it ignores the correlation between features.Therefore,there is still some redundant information in the feature subset resulting in convergence,which will not only affect the efficiency of the algorithm,but also increase the learning cost of the learning model.Therefore,this paper proposes a selection method of the optimal feature subset based on the improved self-tuning adaptive genetic algorithm,abbreviated as ISTA?GAFSS.This method mainly optimizes the feature subset obtained by the convergence of WKNNSTA?GAFS method.The symmetric uncertainty metric is introduced to evaluate the redundant features,analyze the redundancy of the features,eliminate the redundant features in the feature subset,and finally get a feature subset with higher recognition ability.In order to verify the rationality and effectiveness of this method,this method is compared with other 7 feature selection methods on 12 data sets.Experimental results show that this method can achieve higher classification performance with smaller feature subsets.