Study On The Software Quality Prediction Model Based On Artificial Neural Network

With the development of computer system, computers are now used moreand more widely in every aspect and the computer software systems arebecoming more complicated. Therefore, the software reliability attracts lots ofconcern and attention from increasing number of people. How to test thesoftware reliability? It is to appraise and predict the behaviors of the softwarebased on the software reliability model.At present, a lot of software reliability models are built based on thehypotheses of mistake characters and malfunction random behaviors that are leftin software. These hypotheses vary from each other, therefore the difference ofmalfunction behavior types expressed by various software reliability models isobvious; and the testing results are very different from each other. Thus,developing a universal prediction model has become one of the important tasks,which is one of the hotspots in software reliability prediction model. Thesoftware reliability prediction model can predict the software reliability anddoesn't need any premise or hypothesis using neural network.There are two main achievements in software reliability prediction modelusing artificial neural network: firstly, people take the invalid software time as the input of software reliability prediction model; secondly, people use thesoftware quality metrics as the input of neural networks in the software qualityprediction model. This paper proposes two software quality prediction modelsusing software quality metrics as the input of neural networks: the first one isthe software module risk predition model based on LVQ (Learning VectorQuantization) Neural Network. LVQ neural network doesn't need to adjust allthe weights and it has a good stability. Moreover, LVQ neural network has aperfect function of supervised learning and gives classified information asguiding signal, so it can increase the precision of the software module riskprediction. The experiments results show that this model has improved 1 timeand 4 times in the predictive accuracy of high risk module and low risk modulethan the software module risk prediction based on BP neural network. The otheris the software quality prediction model Based on PCA-Wavelet NeuralNetwork. This model uses Principal Component Analysis (PCA) to eliminatethe multicollinearitiy of the experimental data. The PCA can reduce dimensionsof the data, but it remains 97％of the original information. In addition, thismodel applies genetic algorithm to optimize wavelet neural network that canimprove the prediction ability of WNN, and the software quality predictionprecision can be correspondingly improved. The experiments results show thatthis model can achieve a preciser accuracy than the software quality predictionmodel based on neural network of Back Propagation (BP) and GeneralizedRegression Neural Network (GRNN).