Software Defect Prediction Based On Neural Network

The popularity of software promoted the rapid development of society, economy and national. At the same time, this phenomenon also caused many inconvenience to people and even endanger the national security. As we can know from the examples, the main reason to this phenomenon is software error which caused by software defects. Only analyze the quality of the software and predict software defects in advance can we reduce the probability of software error.Software defect prediction technology is a process on the basis of learning the historical data by establishing the defect prediction model, which contributes to make empirical feedback for inner software modules, producing the impact of loads for software quality. At present, the software defect prediction model based on neural network is a hot research topic in this field. However, the use of neural network to predict mainly exists in the following two aspects of the problem: On the one hand, due to the excessive and more complex metric unit, therefore when using the neural network to predict the software defect, it can't get the desired effect; On the other hand, because of containing many kinds of algorithms based on neural network, what kind of neural network algorithm is used to predict the optimal result of software defect prediction, which is also concerned by the relevant scholars. Aiming at these two problems, this paper mainly carries out the study from the following 3 aspects:(1)MDP NASA(National Aeronautics and Space Administration Metric Database Program) data focus on the problem of high dimension of software metrics and data duplication, these problems can lead to low accuracy of prediction results and the result of excessive evaluation. In order to improve the accuracy of prediction and solve the problem of over evaluation, this study uses the principal component analysis algorithm to select the best feature of the MDP data set to achieve the goal of feature dimension reduction.(2)The software defect prediction theory and prediction method based on neural network algorithm of defects were studied. Analysis the network structure and the algorithm flow of the BP(Back Propagation) Network neural network, the Elman neural network and RBF(Radial Basis Function) neural network, and the above three kinds of neural network algorithm to construct software defect prediction model. Through three kinds of software defect prediction model in the MDP data prediction results are comprehensive performance comparison and analysis, selected for software defect prediction of optimal neural network algorithm.(3)In order to verify the feasibility of the defect prediction model, In this paper, first, This study to collect related defect measurement data from a practical software system, Then, classifying these metrics by defect prediction model, the classification results determine whether the software has a defect; Finally, this experiment will one by one to find the forecast for defective modules, and verified the accuracy of the prediction results of the model.The results of this research shows, that the results of principal component analysis algorithm can effectively the software metrics is used to reduce the dimensionality; defects prediction using BP neural network algorithm. The proposed method can effectively to software defect prediction.