| With the development of science and technology, the software systems are more and more complex and the function is more and more powerful, so the demands of software quality are more and more high. The software quality directly impacts the use and maintenance of software, which is impacted by the uncertain facts in the process of software development, and this brings a lot of difficulties to the software quality evaluation. If we can early get the necessary levels in the early software development process, this will be great significance for the quality controlling of the achieving the ultimate software, shortening the software development cycle, reducing the cost of the development and maintenance of software.The technology of software quality prediction model is the key technology of software quality evaluation, but the models which current technology of software quality prediction are based on are relatively rough, and the methods used are mostly statistical algorithm So how to choose the suitable method which can make the software quality prediction model accurately and effectively establish the uncertainty and nonlinear relationship between internal properties and external properties is the concern research topic.For these problems, this paper researched on the software quality prediction model based on software quality metrics. Applying the nonlinear approximation ability and learning and adaptive ability of neural networks, this paper used the software quality metrics for the inputs of neural network and proposed two kinds of software quality prediction models. One is a kind of software quality model based on fuzzy neural network applying rough set, which first used SOM network to discrete data, then used attribute reduction algorithm of rough set to extract the streamline rules in the samples, and finally constructed the fuzzy neural network based on extracted rules. This model simplified the structure of the neural network model, shortened the training time, and improved the predicting ability of model. Another is a kind of software quality model based on the generalized dynamic fuzzy neural network, which used fuzzy-completeness as the determine criterion of the width of Gaussian function that avoid the random of selection in the initialized process. At the same time, this model can evaluate the fuzzy rules and the importance of input variables, so the width of input variables of each rule can be adjusted by its contribution to the implementation of system. The model had an outstanding advantage in learning efficiency and identifying accuracy.Finally, this paper used the simulation experiments of two models to train the performance of model. The results of experimental show that the two models had good predictive ability of software quality, which can better show the nonlinear relationship between software quality metrics and software quality, and had higher prediction accuracy. |
