Research And Improvement Of Feature Selection Method Based On Crow Search Algorithm

In the era of big data,it is a very challenging task to obtain usable information from huge and complex data sets,which makes the field of data mining one of the key areas that scientists pay attention to.For data mining algorithms,obtaining a usable data set is often the first step of the algorithm.When the dimension of the data set becomes high and there are a large number of redundant and useless features,data analysis will become extremely difficult,and the performance of subsequent algorithms will also be greatly affected,so data preprocessing is very necessary.Feature selection is an important method in data preprocessing.This method selects relevant and non-redundant features from a large number of features,and makes the processed data set not affect the classification accuracy to the greatest extent,so as to achieve the purpose of reducing the dimension and improving the accuracy of subsequent classification algorithms.When the search space becomes very large,feature selection can be regarded as an optimization problem.In fact,in a data set with N features,there are 2schemes for selecting the optimal feature subset,so when the number of features N is too large,the exhaustive search method is not feasible.To solve this problem,meta-heuristics have been introduced to solve the feature selection problem with good results.The Crow Search Algorithm(CSA)is a recent algorithm proposed according to the living habits of crow groups.It is a meta-heuristic algorithm based on population.Because of its simplicity and easy implementation,it has attracted the attention of scholars.However,like other meta-heuristic algorithms,it has the disadvantage of slow convergence and easy to fall into local optimum,which seriously affects the classification accuracy and convergence speed of the algorithm.Therefore,this paper first discretizes CSA to form BCSA(Binary crow search algorithm,BCSA)algorithm,and then improves BCSA,and proposes a new feature selection algorithm BICSA(Binary improved crow search algorithm,BICSA),Finally,experiments are carried out to demonstrate the excellent performance of BICSA in solving the feature selection problem.BCSA can be divided into two stages:initialization and update.In this paper,three operators are applied to improve the algorithm for these two stages,improve the accuracy of the algorithm,and speed up the convergence speed.In the initial stage,the randomly generated population in the original algorithm was replaced by a chaotic map,which was chosen because the chaotic map has the characteristics of randomness,ergodicity and dynamic behavior,which can generate a well-distributed feasible solution space,thus avoiding the disadvantage of randomly generating populations.Then,the Opposition-Based Learning(OBL)method is used to calculate the opposite population of the chaotic population.Compared with the randomly generated population at the initial stage of the algorithm,the generation of the opposite population often enables the algorithm to find the optimal solution with a greater probability.Therefore,the addition of OBL enables the initial population to approach the optimal solution from multiple directions,increasing the possibility of the algorithm finding the optimal solution.Finally,according to the fitness function value,the solution with high fitness value is selected from the chaotic population and the opposite population.The addition of these two methods enables the algorithm to obtain an initial population with better quality in the initialization stage,thereby accelerating the convergence speed of the algorithm;In the update phase,the Differential Evolution(DE)algorithm is introduced to improve the performance of the algorithm by using a hybrid algorithm of BCSA and DE.First,BCSA is used to search the feature space,and then DE is used to mutate,cross,and select the generated solutions to obtain better global solutions.The addition of the DE operator expands the range of generated solutions,enabling the algorithm to perform better in the search space.A comprehensive search reduces the algorithm trapped in a certain search area to a certain extent,making it easier for the algorithm to find the optimal solution.These three operators make the improved algorithm have faster convergence speed and more likely to obtain the global optimal solution.In order to demonstrate the performance of BICSA algorithm,this paper compares and analyzes feature selection algorithms proposed in recent years on 16 datasets.The comparison results prove that the improved algorithm BICSA has higher classification accuracy and higher dimensional compression ability.