The Research Of Software Defect Prediction Model Based On Support Vector Machine Optimized By Genetic Algorithm

Software defect is a problem or an error in computer software which causes the software can not run normally, or a hidden functional defect. With the continuous development of computer technology, the scale and complexity of software systems are growing exponentially. Software testing and defect repairing always require a lot of manpower and time cost because of the complex software structure and the large amount of code. Software defect prediction aims to use special methods to find which modules of software system may be defective or to predict the number and distribution of the defects in the software system. This can provide useful guidelines for software testing.In the software defect prediction research, the method which establishes the software defect prediction model using support vector machine with nonlinear calculation ability to analyze the software metrics and defect data has been proved to be more effective. However, support vector machine lacks unified and effective parameter optimization method. To solve this problem, this paper proposes a software defect prediction model based on support vector machine optimized by genetic algorithm making use of its global search ability and potential parallelism. This model uses support vector machine as the classifier of software defect prediction and uses genetic algorithm to select the optimal metric attributes and to calculate the optimal parameter of support vector machine. Compared with principal component analysis, this method can avoid premature useful information screening when selecting software metrics and further improve the accuracy of software defect prediction. In addition, this paper proposes to set the cross probability and mutation probability of genetic algorithm dynamicly in order to avoid premature convergence and falling into the local optimal.In order to verify the validity of the method, this paper uses NASA MDP data sets to experiment. We determined the parameter of fitness function by comparing different average maximum fitness. Compared with traditional genetic algorithm, improved genetic algorithm proved its validity and superiority, because its probability of convergence and average convergence generation were both better than traditional genetic algorithm’s. Finally, we predicted defects of different software projects and analyzed the experimental data carefully. The experimental results showed that the prediction accuracy, precision, recall and other performance indicators of this model are better than other models.