Research And Application Of Feature Selection Based On Modified Binary Cuckoo Algorithm

With the widespread attention and extensive research on feature selection algorithms,the filtering,encapsulation and their hybrid models in feature selection have been widely used in different fields.In recent years,the heuristic algorithm-based encapsulation model and hybrid filter package model have gradually become the research hotspot in the feature selection algorithm.Because of its easy implementation,the cuckoo search algorithm has fewer parameters,the computational cost is relatively low,achieve convergence effect and much attention.In this paper,the feature selection algorithm has been mainly studied.For the binary cuckoo algorithm in the heuristic algorithm involved in the hybrid filter package model,an improved version of Modified Binary Cuckoo Search(MBCS)combined with the feature-based hybrid filter package model,MBFS-based hybrid filter package feature selection algorithm is applied to the classification of mammograms.The main research contents are as follows:(1)Propose a Modified binary cuckoo algorithm.In order to further reduce the running time of the binary cuckoo algorithm and improve the efficiency of the algorithm,the article mainly changes the coefficient of the mapping function to a change value to achieve an adaptive effect.Secondly,using the genetic algorithm for mutation Probability of sub-substitution in the original algorithm.By experimenting on the common 0-1 knapsack dataset and comparing with the existing algorithms,the improved method further improves the speed of the algorithm,and the performance of jumping out of the local optimum is also improved.(2)Propose the MBCS based hybrid filtering package feature selection algorithm model.The filtering algorithm directly evaluates the characteristics of the whole data set,and runs fast,but the classification accuracy is relatively low.The encapsulation algorithm uses the training accuracy of the subsequent learning algorithm to evaluate the feature subset,and the classification accuracy is relatively high,but the computation is relatively large,Not suitable for large-scale data sets.In this paper,the MBCS is introduced into the hybrid filter package model of feature selection.The experimental results show that the proposed method preserves the existing high classification performance of the package method and further shortens the computation time.At the same time,the size of the the selected features' set shrinks again.(3)A new medical image classification method based on MBCS-based hybrid filter package feature selection algorithm.In order to expand the application of MBCS-based hybrid filter package feature selection algorithm in the classification of real data set,the MBCS-based hybrid filter package feature selection algorithm is applied to the classification of breast light image data set.The original source of the dataset is the Mammography Image Analysis Society(MIAS),which is a commonly used standard dataset of mammography images.After the preprocessing of the original image,the grayscale co-occurrence matrix method is used to extract the image texture features to construct the feature dataset.Hybrid filtering package feature selection algorithm for feature selection of the feature set obtained by the experiment and the other three feature selection algorithm contrast average classification accuracy can be seen that the proposed MBFS-based hybrid filter package feature selection method can MIAS data set Get the ideal classification results.