Technology Research And Application Of Software Defect Prediction Based On Bayesian Networks

Research indicates that in the development life cycle it is better to find defects earlier. The defects can be corrected in time. Thus can avoid a great deal of workload of modifying, remedying and correcting the defects in behind development process and can reduce customer complaint. It can shorten development cycle, improve software quality and the contentment degree of consumer, and reduce cost. So this paper mainly considers every phase of software development lift cycle, analyze correlative factors of generating defects of every phase and process and correlation of defects factors. And this paper recurs to the establishment of defects warehouse and management technique, to manage efficiently and analyze defects deeply, accumulate team development software defects knowledge and the defects, to inherit these experience and knowledge and predict intelligently. On the basis of the before mentioned, a kind of advanced, effective and simple defect predict method and technology based on Bayesian network is proposed. It can predict defects effectively and somewhat accurately and can help to predict and find defects earlier and effectively, then to achieve the aim of correcting defects.For this purpose, the theory and method of software management and software defects classification methods are studied firstly. An improved model named software defects management model based on software development process is presented. On this base, software defects generate factors and their relationship are studied and analyzed in detail. The factors resulting in the software defects are analyzed through Bayesian network method. Bayesian network fabric of requirement analysis phase, design phase, coding phase and testing phase is established, and at same time it is analyzed. Then, on the basis of analysis and Comparing software defects prediction model and considering that the software prediction model should not only have the capability of dealing with multiple complex factors and accumulating experience data, but also provide the function of flexibility in construction. Combine Bayesian with prediction model, a software defects prediction model based on Bayesian network is introduced, and the computing process of this model is detailed, and the impact-degree of every factor that lead to result is analyzed. To compute conveniently, this arithmetic is implemented in JBuilder2005, arithmetic flow chart and pseudo-code are introduced in this paper. At last this theory is applied to a specific example of software requirement analysis phase of simplified predict model. The calculating result is analyzed and the validity of model is validated.