Improving Crow Search Algorithm To Optimize KNN Parameters And Feature Selection For Vulnerability Classification

There are many optimization problems in social production and life.Optimization problems are often complex,high-dimensional,uncertain and non-linear.Solving optimization problems is one of the hot areas of current research.Meta-heuristic algorithm solves optimization problems by simulating life phenomena in nature.It has the characteristics of simple operation,strong parallelism and good optimization effect.Meta-heuristic algorithm has applied to optimize the parameter in machine learning classification model to improve the classification performance.In addition,meta-heuristic algorithm is implyed to select the optimal feature subset of the classification data set to further raise the classification accuracy.In this paper,meta-heuristic algorithm will be used to optimize the parameters and features selection in the classification model.The optimized classification model will be applied to vulnerability classification.The main work of this paper is the following three aspects.Crow search algorithm(CSA)is a meta-heuristic algorithm proposed by simulating crow foraging behavior in 2016.CSA has the advantages of simple structure and easy implementation.At the beginning of iteration,the population of CSA is single and unevenly distributed,so the algorithm is easy to fall into local optimum and appears low accuracy.The improved crow search algorithm based on dynamic parameters and Gaussian mutation(ICSA)is proposed to solve these problems.The parameter dynamic change and Gauss mutation are introduced in the process of iteration so that the larger parameter value and mutation space are used in the initial iteration stage to improve the diversity of solution and avoid CSA falling into local optimum.At the later stage of iteration,the parameter value and the Gauss variation space become smaller,so that CSA can accelerate the convergence speed and increase the precision.ICSA is intruduced to optimize the parameters and feature selection of KNN classification model.KNN can obtain better parameters by ICSA parameter optimization than random selection.ICSA is applied to select a subset of classification features so that the dimension and computational complexity of data can reduce in KNN classification.The combination of ICSA parameter optimization and feature selection can improve the accuracy of KNN.KNN which is optimized the parameters and features selection by ICSA is employed to classify vulnerability.Test are conduct on the vulnerability data sets of CNNVD and NVD.The test result prove that the optimized KNN can more effectively simplify the characteristics of vulnerability data sets and improve the accuracy and efficiency of vulnerability classification.