Abstract:

The prediction of software defects has been an important research topic in the field of software engineering. The paper focuses on the problem of defect prediction. A classification model for predicting software defects based on the integration of rough sets and support vector machine model (RS-SVM) is constructed. Rough sets work as a preprocessor in order to remove redundant information and reduce data dimensionality before the sample data are processed by support vector machine. As a solution to the difficulty of choosing parameters, the particle swarm optimization algorithm is used to choose the parameters of support vector machines. The experimental data are from the open source NASA datasets. The dimensions of the original data sets are reduced from 21 to 5 by rough sets. Experimental results indicate that the prediction performances of Bayes classifier, CART tree, RBF neural network and RS-SVM are all improved after the dimension of the original data sets are reduced from 21 to 5 by rough sets. Compared with the above three models, RS-SVM has a higher prediction performance.

Key words: rough sets;support vector machine;software defect;prediction;particle swarm optimization