Research On Software Defect Prediction Based On Principal Component Analysis

Finding and fixing software defects has been one of the most expensive activities of software development and maintenance,so the software defect prediction is an important research topic in software engineering field,especially for solving inefficient software testing and review method in the existing industry.Accurate prediction of defect-prone module can help software engineers to allocate the limited resources to those modules which are most likely have defects in the software testing and maintenance phase,thus improving test efficiency and reducing the maintenance cost of software.In this paper,the main research contents include the following three aspects:Firstly,software defect prediction model based on principal component distribution function.In order to solve the data-imbalance problem in the defect prediction,this model uses the distribution functions of principal component to generate random numbers and combines with the random under-sampling.At last,the feasibility and effectiveness of this model are verified by comparing the experimental results.Secondly,software defect prediction model based on weighted kernel principal component analysis.The model uses the weighted kernel principal component analysis instead of principal component analysis to solve the problem that the dimension reduction result of principal component analysis is not good.The model also uses Tomek under-sampling method to replace the original random under-sampling,which can prevent the loss of important information.The effectiveness of model is illustrated by comparing experimental results.Thirdly,PCA-Copula software defect prediction model based on the principal component analysis.The PCA-Copula model is proposed to solve the problem that it is difficult to determine the suitable model for software defect prediction.At last,the PCA-Copula model is compared with the multiple linear regression,logistic regression and neural network to show it's validity and feasibility.Finally,the results of the three defect prediction models are summarized,and the future development is prospected.