Software Defect Prediction Based On Baysian Network

With the development of the Internet， the software industry is developing rapidly.Software bring much convenience to people，at the same time，defects also bing a lot oftrouble. Some related investigation and study showed that： the infinite accumulation ofsoftware defects is the key factor that leading to a series of problems， and the timing ofthe software defects as early as possible， if discovered early. To reasonably control thedevelopment costs in the software quality assurance at the same time， it must beobserved， statistics， analysis of the former defects， and to summarize the regularity andforecast the next.Predictably， there are many factors that can lead to defects， coupled with thelimitations of the complexity of the project itself and the test method， and the uncertaintyof knowledge， makes the structure of software defect prediction method is not an easything. The knowledge of bayesian network to solve the uncertainty problem has manyadvantages， so it becomes the most popular and the most ideal prediction model. But it isnot perfect， it will become the NP hard problem when there are many factors leading todefects.For the above， this thesis’s work mainly include the following aspects：(1)Research and elaborates relevant theoretical knowledge of software defects，including defects definition， classification， causes， and then the common defectsprediction model were analyzed and compared， including Matrix Data Analysis Chart，Artificial Neural Networks， Control chart method， capture model method. Thus， theBayesian networks model， which lays the foundation for follow-up study.(2)Summarize the theory knowledge of Bayesian network， including the concept，representation and inference process， On this basis， it focuses on the NP-hard that causedby the increasing relevant factors in the practical application， and proposedimprovements. Mainly through a combination of Bayesian independent nature of thenetwork itself， the original Bayesian network can be simplified. Finally， the improvedalgorithm formula as well as before and after the time-consuming comparison， as thetheoretical basis in fact.(3)Applied the improved Bayesian network to specific project examples， with the combination of historical data and expert experience predicting software defect， it isobvious that greatly simplifies the calculation resulting in improved operational efficiency，and analysis the experimental results and evaluate the predictive validity of the model.