Research On The Application Of Improved Sparrow Search Algorithm In Feature Selection

The "Curse of dimensions" caused by the rapid development of information science has negatively affected the performance of machine learning models.Feature selection provides a method to identify essential features and remove irrelevant or weakly relevant features from the dataset.Evolutionary algorithms have gained more and more attention in dealing with feature selection problems.Many scholars have proposed various optimization algorithms to find a better algorithm to solve these complex optimization problems more efficiently.However,some defects still exist,such as a slow convergence rate,non-convergence,and quickly falling into local optimization.To solve these problems,the main work of this thesis is as follows:1.As the Sparrow Search Algorithm is prone to fall into local optimization due to poor population diversity,this thesis proposes a new algorithm called TFSSA.First,the improved chaotic map is used to initialize the algorithm to increase the ergodicity and diversity of the population;Secondly,LFs are used to expand the search scope,help the algorithm find the search location more effectively,and then improve its global search capability;Then,expand the search scope of sparrows by updating the location of producers and adjusting the number of patrolmen;Finally,to further improve the convergence accuracy,the optimal individual is mutated.This thesis uses the classical test function set CEC2020 to verify the algorithm’s performance and compares it with seven comparison algorithms.The results show that combining these strategies can improve the sparrow search algorithm’s performance.2.The TFSSA is used to select the best feature combination,thereby maximizing the classification accuracy and minimizing the number of selected features.This thesis compares TFSSA and nine different methods on 21 datasets.In addition,the method was applied to the data set of COVID-19,and the average classification accuracy of 93.47% and the average feature selection number of 2.1 were obtained,respectively.Experimental results show that the algorithm has certain advantages in improving classification accuracy and reducing the number of selected features.