Study On Software Defect Prediction Based On Multi-gene Genetic Programming

Software defect prediction refers to the construction of a trained model by mining software historical repositories,and then using trained models to predict defects in new software modules.Most current software defect prediction models aim to predict the number of defects in a given software.However,it is difficult to accurately predict the number of defects due to the noise data in the defect data.The researchers further propose to rank the software modules by predicting the relative number of defects.In addition,due to high dimensions and the correlation between the features of the software defect dataset,there are some redundant features.It is necessary to perform feature selection or feature dimension reduction before constructing the defect prediction model,which makes some useful information be deleted prematurely,thus reducing the performance of prediction model.This paper analyzes the relationship between the features of each module and the amount of the defects in the software defect prediction.Firstly,we study on the optimization of the non-linear prediction model based on the genetic algorithm according to the idea of directly optimizing the ranking performance of model,and then considering the advantage of Multi-Gene Genetic Programming(MGGP)in dealing with multicollinearity problems,we use MGGP to build a software defect-ranking model,and the relative ranking of defects in software modules is predicted and evaluated.In order to verify the predictability of the software defect-ranking model,we perform experiments on 11 publicly available defect datasets,and the Fault-Percentile-Average(FPA)is used to evaluate the performance of the MGGP-based and other software defect prediction methods.The results show that the software defect prediction model constructed by MGGP is superior to the existing nonlinear prediction models in defect ranking.The.nonlinear prediction model based on genetic algorithm is superior to the original nonlinear prediction model.In addition,the software defect-ranking model constructed by MGGP method can use the original features to construct a prediction model with better performance without considering the influence of the correlation between the software module features.