Research Of Data-driven Software Reliability Combination Model

With the rapid development of computer technology in recent years, software reliability gets more and more attention. High stability has been the goals of software system, while the stability depends on the software reliability, which will cause a great loss if it is not guaranteed. Thus, it is very important to build software reliability model to predict the software reliability.In the aspect of software reliability prediction model, some scholars have carried on a lot of fruitful research, including the traditional linear regression model and modern intelligent algorithm, etc. Nevertheless, the selection of model inputs is too subjective in the past study of software reliability model, and mostly it ignored the influence of model inputs to the software reliability prediction accuracy. Furthermore, hypotheses exist when analytical models are used and the single model cannot describe the change rule of the software reliability, which result in the low prediction accuracy and less commonality.To solve the above problems, the paper focused on the software reliability and put foreword the data-driven software reliability combination model. The emphasis are on the input selection of data-driven software reliability model and how to predict software reliability by combining multiple software reliability models, which can improve the reliability prediction accuracy and universality extremely. Firstly, we have studied the time-series data preprocessing and proposed an input selection method of the software reliability model based on sliding window mechanism and AGA. In addition, to solve the problem of weights dynamic changes in the process of constructing combination model, the fuzzy adaptive variable weight method was put forward. Then, according to the model input selection techniques and fuzzy adaptive variable weight determination method mentioned before, the paper combined different data-driven software reliability models. The research will be beneficial to improve the software reliability prediction accuracy and be helpful to measure and forecast the features of software product effectively, and then the conditions during the software development will be better controlled.Lastly, the simulation experiment was carried out in view of the model built in this paper. The results turn out that, the data-driven software reliability combination model can improve prediction accuracy and versatility and can provide a powerful technical support for guaranteeing the software quality and developing reliable software.