Research On Static Multi-objective Software Defect Prediction Strategy

Software defect prediction is a technology that predicts the future development of software defects according to the historical data and defects that have been found.This paper has done the following work for software defect prediction:First of all,for the two targets of probability of detection and probability of false alarm,the existing research mainly focuses on multi-objective particle swarm optimization algorithm.The algorithm generates rules of the classifier contains multiple rules,each rule is not uniform and there is a duplicate area coverage.Generated rules need to be combined to predict software defects.In this paper,the multi-objective oriented cuckoo algorithm is introduced to optimize the parameters of the prediction model.The individual position of the cuckoo algorithm is the optimal position of the individual history.This feature makes the distribution of the rules more uniform and improves the performance of the algorithm.In order to verify the algorithm,eight data sets of the NASA database are selected and compared with the results of the eight algorithms.The experimental results show that the average performance of the algorithm is superior.Secondly,we analyze the distribution of the number of lines of code in the six data sets of the NASA database.The results show that most of the software modules have fewer code lines(called small modules)and only a small number of modules have a long line of code(called a large module).According to Arisholm's research conclusion: the test resource is proportional to the number of lines of the software module and the probability of false alarm of the large module in the defect prediction process will seriously waste the test resources.In order to reduce the waste of resources due to large module false alarm,we divided a complete data set into large modules and small modules according to experience,and each assigned a support vector classifier to achieve the goal of reducing the overall waste of the test resources.In order to further verify the technical performance,this paper compares the algorithm with nine algorithms,and the results show the effectiveness of software defect prediction based on multi-objective two support vector machine.Finally,the multi-objective defect prediction based on two support vector machine needs to divide a data set into large modules and small modules according to the number of lines of code.The choice of this ratio has great influence on the effect of defect prediction.So we use the golden section of the five data sets of the division ratio made a choice.The experimental results show that the ratio is between 40%-80%,and the effect is better.