Research On Software Defect Prediction Model Based On SAPSO-BP

The number of software defects is an important index to reflect the software quality. If software defects are found as early as possible, it can help us avoid the waste of resources on modifying and making up the defects in the late period of software development, thus ensure the software delivery. Software defect prediction models can be used to predict the distribution or the number of defects at the early stage of software development. Software defect prediction technology makes the software developers focus their limited resources on the high incidence of defect modules, which can be more effective to find and eliminate these defects, to meet the demands of users. It plays a very important role in software’s reliability improvement and quality assurance.Since software defect prediction technology was proposed by Briand in1992, it has received great attention in the field of software engineering until now. There are many researchers researching on it, they have put forward a lot of software defect prediction technologies. BP Artificial neural network is one of the defect prediction techniques, it has been widely used and has high performance. This method makes full use of the attributes’value of software module and the past defect data to give an analysis. Then it predicts on the classification and the number of the defects. Although this model has achieved a good prediction effect, due to some problems lying in the BP algorithm, the accuracy of defect prediction is still not ideal.In order to solve the issue, this paper firstly studies the software defect prediction model based on BP neural network, and proposes the software defect prediction model based on improved SAPSO-BP neural network to improve the ability of software defect prediction. In this paper, our work includes the following aspects:Firstly, this paper studies the theory of software defects and software defect prediction technology, it includes the measure properties describe defects, the steps of creating software defect prediction models, analysis and comparison of common software defect prediction models and so on. The above work provides a valuable reference for the research of software defect prediction model based on SAPSO-BP network.Secondly, this paper studies the software defect prediction model based on BP network and point out the deficiencies in the present model. In the traditional model, based on BP algorithm, it uses gradient descent method to adjust the weights and threshold values, which makes the algorithm has a low convergence and easily fall into local optimization. These problems lead to the prediction rate with low accuracy.Thirdly, for the shortcomings of the traditional model, the paper puts forward the software defect prediction model based on SAPSO-BP network. In order to solve the problems in the traditional BP algorithm, we employ particle swarm algorithm to initialize parameters of BP network. Next, we take the advantage of the global optimization ability in the simulated annealing algorithm to fix the weights and thresholds in the network. By this way, we can solve the problem of falling in a local optimum. Finally, we use the trained network model to predict the software defects.Lastly, the paper make use of the data sets of MDP project from NASA website to predict software defects using the software defect prediction models based on the traditional BP algorithm and the improved SAPSO-BP algorithm, respectively. The experimental results show that the software defect prediction model based on SAPSO-BP neural network has a high prediction rate.