Research On Software Defect Data Prediction Based On Data Mining

Software defect is an inherent property of the software, its main hazards are reducing thereliability of software, increasing development cost and extending the development cycle.Software testing is a very effective tool to find software defect timely and improve softwarereliability. While accurately predicting the distribution of software defects have a greatsignificance for the software testing process. With the rapid development of computer technology,software size and complexity is growing exponentially. To predict the software defects accurately,people have to take more complicated impact factors into account. At this time, the traditionalprediction models are difficult to deal with the reasoning prediction which has complex causalrelationship and uncertainty knowledge, and the prediction results of these methods are often toobroad to lose its practical significance. To solve this problem, people try to apply research methodsof other fields into software defect prediction, data mining techniques are common methods.Data mining is a new research topic in the field of database system. It can analyze mass dataand extract useful knowledge from them to provide effective basis for decision-making. Thisthesis is on application of data mining in predicting static software defect and dynamic softwaredefect. The main work and contribution of this thesis are summarized as follows:Static software defect detection aims to automatically identify defective software modulesthat accelerates efficiently software test and improves the quality of a software system. Due to theapplication of traditional software prediction model being limited for its low accuracy andapplicability, this paper puts forward a software prediction model based on PSO-BP, whichemploys Particle Swarm Optimization (PSO) to optimize weight and threshold value of BP. Weuse cross-validation method as experiment method, and compare the result with other machinelearning methods-BP and J48. The result shows that PSO-BP has higher prediction accuracy. AndWe also talk about how to set the parameters of PSO.Dynamic software defect prediction results can be timely feedback to the testers, and improvethe software testing process. We put forward the ESGM model because the traditional dynamicdefect prediction models assume too much and versatility of model is weak. The model uses theEMD algorithm to decompose the original defect data sequence, then predicting on IMF andremainder by PSO-SVR and grey theory. The model is applied to the data set SYS1.The resultsshow that the model ESGM fully extract the advantages of grey model and PSO-SVR model, themodel can get better prediction, and the model can be used in dynamic software defect prediction.