Improved Fireworks Algorithm And Its Applications On Joint Optimization Of Feature Selection And SVM Parameters

Firework algorithm is a swarm intelligence algorithm emerging in recent years.The firework algorithm has the ability of self-regulation between global search(exploration)and local search(exploitation),and has a strong ability to solve complex optimization problems,which has attracted widespread attention from researchers.Support vector machine(SVM)is a common classifier in the field of machine learning,which can be applied to non-linear and high-dimensional problems,and is particularly suitable for solving small sample classification problems.The value of parameters in SVM greatly affects the classification performance of SVM,and good feature selection methods can effectively remove redundant features and improve the efficiency of classification.Based on the above background,this paper studies the firework algorithm and its application in joint selection of feature selection and SVM parameters.The main work is as follows:Firstly,this paper proposes an enhanced firework algorithm using a loser-out tournament strategy.This algorithm uses a new type of mapping rule that takes into account the location characteristics.It maps explosive sparks that exceed the upper boundary of the explosive space to the area near the upper boundary of the explosive space,and the explosion sparks below the lower boundary of the explosion space to the area near the lower boundary of the explosion space.The improved mapping rule retains the relative position information of the generated spark and the boundary,so that the sparks beyond the boundary can be mapped more specifically.The proposed algorithm introduces a strategy of adaptively adjusting the number of explosive spark parameters to better balance the global and local search capabilities of the algorithm.The 28 functions in the CEC2013 standard test function set are selected for testing.The experimental results show that the proposed algorithm has better search performance.Secondly,a joint optimization method based on feature selection and SVM parameters of improved firework algorithm is proposed.According to the characteristics of this application problem,a new fitness evaluation method is designed.In addition to improving the accuracy of SVM classification and reducing the number of redundant features of the sample,the method can adaptively increase the punishment range of fitness with the increase of the selected feature number,and ensure the diversity and wide range of the fitness value distribution of the feature selection part.In view of the lack of information interaction among individuals in the basic fireworks algorithm,differential mutation operator is introduced to improve the information interaction ability of the algorithm and enhance the local search performance of the algorithm.In the basic firework algorithm,the selection strategy needs to calculate the Euclidean distance between any two sparks in the candidate pool,thereby increasing the time complexity of the algorithm.In view of the above shortcomings,this paper proposes a roulette selection strategy based on fitness values.Compared with the basic firework algorithm,the time complexity of the algorithm is reduced,and the operation efficiency of the algorithm is improved.Nine UCI classification datasets were selected as test sets to verify the effectiveness of the two improved strategies in improving the performance of the algorithm,and the proposed algorithm was compared with the classic genetic algorithm,particle swarm algorithm and the better groups in recent years.Comparing intelligent algorithms,the experimental results show that the algorithm in this paper can obtain higher classification accuracy with fewer feature numbers on most test data sets,and performs best.Thirdly,in views of the method of feature selection and SVM parameter joint optimization based on the improved fireworks algorithm,a breast cancer classification diagnosis device is proposed.The diagnostic device is divided into six parts: data set input,data preprocessing,feature selection and SVM parameter optimization based on the improved firework algorithm,SVM model training,SVM model testing,and classification result output.On the three different datasets of breast cancer original,diagnosis and prognosis in Wisconsin,USA,the proposed classification diagnostics based on the improved fireworks algorithm was compared with the classification model using traditional evolutionary algorithms and machine learning method.The experimental results show that the proposed diagnostic device as a whole can obtain a high classification accuracy rate,which has a strong practical value.