The Research On Support Vector Machine Based Software Reliability Model

In this paper,the impact on software reliability factors,the failure mechanism to assess the model as well as the necessary foundation of mathematics is study.The classical software reliability growth model,which is used in the industrial application widly,is survey.In a large number of software reliability assessment model,the probability of type of model assumptions are often difficult to meet the existence of defects.Bayesian network model to obtain the existence of a priori knowledge is the difficult problem.Artificial neural network will encounter problems such as:difficult determining structure,lack of or too fit to be together and easy to fall into local minimum points,and so on.In order to make the analysis more accurate and intelligent assessment to the reliability of the software,this article discusses the theories of software reliability and support vector machine,presentes two reliability models:(1).Support vector regression based software reliability growth model(SVRSRG). SVRSRG.which is based on support vector regression of statistical learning theory,can avoid the above-mentioned problem.Considering the failure dataset from a true software project as object of study,the process of establishment of SVRSRG model,the problem of kernel function selection,parameter estimation of the cross-validation & grid search method,is illustrated.The algorithm of SVRSRG model and the methods of implementation is proposed. In order to evaluate the SVR-based software reliability growth model with goodness of fit,a total of three models:Goel-Okumoto model,Jelinski-Moranda model and the proposed SVR-based software reliability growth model SVRSRG,is implementation respectively.The experimental results of these model,is evaluated using both MSE and SCC.The results showed that SVRSRG model prediction has a good ability to verify the validity of the method.(2).Support vector machine Based model for early prediction of software reliability.A large number of studies have shown that software quality and software complexity, development costs and production efficiency is closely related.The complexity of the software causing software errors are the main reason.When the complexity of software over a certain threshold value,the software error will be dramatically rise,and even give rise to the failure of software development.In this paper,the relationship between the reliability and the complexity of software is research experimentally.The complexity include 11 metrics totally:total lines of code including comments,total code lines,total character count,total comments,number of comment characters,number of code characters,Halstead's program length(N),Halstead's estimated program length(N^),Jensen's estimator of program length(NF),McCabe's cyclomatic complexity(V(G)),Belady's bandwidth metric(BW),and so on.Support vector machine based model is set up for early prediction of software reliability.The data set namely MIS is selected for experiment.PCA methods is used to eliminate the impact of correlation.The results of several tests show that prediction accuracy can be achieved about 89%around.This is the verification for the feasibility of support vector machine based software reliability early prediction model.