Research On Software Defect Prediction Based On Classifiers Fusion Technology

Software defect prediction technology has become the research focus of software reliability.Firstly,we should describe and classify the software defect,and then the software defect data in the database are analyzed and summarized for defect prediction.This paper discusses two problems which exist in software defect prediction models: Firstly,mostly software defect prediction model always focus on the single attribute in the feature subset other than the mutual influence of attributes,which inevitably lead to the lower prediction accuracy.Secondly,although the high performance of the single classifier has been more widely used in the industry,some features are less robust.In order to solve the above two problems,combining with the classifier fusion theory and taking the software defects as the research object,this paper firstly proposed the prediction method of software defect based on GA-FM(Genetic Algorithm-Fuzzy Measure)fuzzy integral.The prediction model learn the fuzzy measures of each attribute and the combinations of attributes by using genetic algorithms,and then the model fused each attribute by using Choquet integral algorithm,and lastly,the model output the class label of software module.Secondly,this paper proposed an multiple classifiers fusion model by using the fuzzy integral algorithm.Finally,this paper proposed a dynamic software defect prediction methods based on Elman neural network integration.The method selected individuals by using K-means clustering method and integrated the individual classifiers by using the weighted average method.The experiment results show that the GA-FM fuzzy integral static software defect prediction method and the fusion model based on different classification can significantly improve the accuracy of software defect prediction and help for measuring the software reliability;The dynamic software defect prediction method based on integrated Elman neural network proposed in this paper can effectively control and guidance the software defects that may occur during operation or development of the software.