| By predicting the quality of the software that will be formed in the early stage of development, faults brought in at the phase of design will be found out early in order not leave them in the software product. Furthermore, it will be convenient for designers to adopt appropriate plans according to specific expectations of the target software.However, the traditional prediction models have following shortages: 1) can't express the relationship between attributes and metrics effectively; 2) lack of the ability to process data both qualitatively and quantitatively; 3) not appropriate to the case with uncompleted information.On the one hand, artificial neural network is good at nonlinear approaching and holds the abilities of study and self-adaption, but is poor at knowledge explanation. On the other hand, fuzzy logic lacks the capacities of study and self-adaption, but holds the advantages of reasoning and explanation of uncertain information. According to the foregoing problems, as neural network and fuzzy logic is a complementary pair, in this paper, a model based on fuzzy neural network is presented, which holds the abilities of dealing with incomplete information and multiple patterns of data, as well as making effective description of knowledge. The applications of the proposed model in quality prediction for software product lines and for object-oriented software are studied. Training algorithms of this network model, such as BP algorithm and genetic algorithm have been studied in this paper as well. And the performance of each algorithm is obtained through the simulation with MATLAB.The result of the simulation justifies that the prediction model based on neural fuzzy network can not only realize the prediction of software quality, but also implement it under the condition of both fuzzy and accurate data, as well as incomplete information. In addition, the uncertain causal relationships between software quality attributes and inner factors related are also effectively described. |
PDF Full Text Request