Research On Software Reliability Model With Change Points

The prediction and analysis of software reliability is to use the software reliability models for software reliability measurement and forecasting. And its core and key is the software reliability model, which has the important significance for software product quality measurement and evaluation.Software reliability modeling is mainly based on the software failure data while the data is obtained from software testing. In software testing process, the distribution of failure data may be influenced by many factors, such as environmental factors, human factors and so on. Once these factors change,this will lead to a change in the statistical law of the software failure data, which is called “change point”. So the single software reliability model could not accurately describe the failure behavior of the software. It is necessary to introduce the change point analysis into the software reliability modeling. In this paper, three kinds of software reliability models with change points are proposed,including the single or multiple change points. This paper has the following three innovations:Firstly, software reliability model with the change point based on the generalized likelihood ratio(GLR). By a further study of the change point detection method with GLR, we apply it to the software reliability model. Then the model is verified by experimental data of ever literature with change point.Secondly, the hidden Markov software reliability model with EM method. The model describes the change of the software failure behavior by using the hidden state process. By the way, it assumes that the failure data obeys exponential distribution. A new EM algorithm based on the forward backward algorithm and Viterbi algorithm is proposed to estimate parameter. It can quickly and effectively detect the change point, and can effectively improve the reliability of the software evaluation and fitting accuracy.Thirdly, the hidden Markov software reliability model with B-splines. To solve the problem that the distribution of the software failure data is difficult to determine in the reliability prediction model of the hidden Markov software, a non-parametric probability density estimation method of B-spline is introduced into the model. This model estimate the failure distribution by B-spline without any assumption on the software failure distribution. It is more universal in practical application.Finally, the model fitting results of the three kinds of software reliability model with change points are summarized, and the future research is prospected.