Classification Optimization Technology Based On Improved Crow Algorithm

In order to solve some of the shortcomings of traditional classification problems: 1)After the extraction of data features,the number of data features dimensions is too high,resulting in curse of dimensionality.It is difficult to extract high-quality feature subsets,and too many redundant features are not conducive to later classification tasks;2)The single classifier is insufficient,the classification effect is unsatisfactory,and the weight adjustment of multiple classifiers is difficult.Therefore,this thesis uses swarm intelligence optimization algorithm-improved crow search algorithm for optimizing machine learning classification technology.The main work is as follows:1.This paper uses the Binary Hybrid Crow Search and Differential Evolution Algorithm(BCSADE)to study the feature selection technique and searching for optimal feature combination on the given data set.The feature selection problem can be regarded as a discrete optimization problem.This paper will improve the crow-differential algorithm for discrete processing.The global optimal particles are quickly found by using the distance differential information of the difference population,then improving the step-length moving mode of crow individuals.Considering the characteristics of crow population with memory matrix,making the best position.Information interaction between crow populations and differential populations is used to increase population diversity to avoid falling too quickly into local optimization.Through collaborative search,the ability of global exploration and local development is balanced.The ultimate goal is to achieve dimensional parsimony and improve the quality of subset features.2.The Cauchy Mutation Crow Search Algorithm with adaptive step size(CMCSA)is used for the study of weight assignment problems in multiple classifier integration techniques.The solving coefficient combination can be seen as a continuous type of optimization problem that searches for the optimal solution over a real number range.This paper improves the algorithm search capability by improving the step-length update and moving strategy of the crow search algorithm,adaptively adjusting the weight combinations according to the target function values,and iteratively searching for the optimal solution,thus improving the classification accuracy of the integrated model.3.The BCSADE algorithm is applied to ten common UCI datasets for feature subset search,and the CMCSA is verified for algorithm validity by using ten benchmark functions.Suitable base classifiers are selected for integration by discrepancy metrics,multiple classifier integration models are constructed using CMCSA for classification learning,and the accuracy of the algorithm is evaluated by 10-fold cross validation.Then the proposed algorithm is used in the web detection domain.The results of the simulation experiments show that the classification correctness of this method is improved compared to the other eight commonly used intelligence algorithms,and it has potential in the classification technique optimization study.