Research Of Software Defect Prediction Model Based On LLE-SVM

Software plays an important role in social life. And the number of software defect is increasing with the increase of software complexity. The harm caused by software defects is becoming more and more awful, and it does damage to the software security as well. The prediction of software defect can improve software security by helping programmers find software defects accurately.The prediction of software defect is a technology which can get the information of defects by analyzing the metric data. So, more attributes of the metric data should be defined in order to describe the software system more comprehensive ly. However, the overmuch attributes will cause the redundancy of data which could decrease the accuracy. To solve this problem, A model based on locally linear embedding and support vector machine(LLE-SVM) for the prediction of software defect is proposed in this paper. In the process of researching, we analyzed the characteristics of software defect data set, using local linear embedding algorithm(LLE) to achieve the reduction of data dimensionality. The software defect data in a high dimensional Euclidean space would be come down to a lower dimensional space, software defect data retention manifold structure in high dimensional Euclidean space, then obtaining the corresponding map, and achieve the reduction of data dimension. Using support vector machine(SVM) as base classifier, and training SVM by data of reduction dimensionality, gives the LLE-SVM software defect prediction model.This paper implements the LLE-SVM software defect prediction prototype system. The comparison between LLE-SVM model and SVM model was experimentally verified as same as the defect data set of NASA. We chose the CM1 data set in the NASA data set fo r experiments. The results indicate that the proposal LLE-SVM model performs better than SVM model, and it is available to avoid the decrease of accuracy caused by the data redundancy.