Software Defect Prediction Based On Automatic Feature Engineering

Early software defect prediction usually used data mining technology.Later,people improved the software defect prediction model in different aspects,such as putting forward new metrics features,optimizing data mining algorithm and so on,to improve the performance of the prediction model.A software defect prediction system based on automatic feature engineering is designed to improve the software defect prediction model from the aspect of feature engineering.The software defect prediction can be completed automatically by realizing the automation of feature engineering.Feature construction and parameter learning are two important stages of feature engineering.In feature construction stage,a combination method based on AdaBoost is proposed to realize feature linear combination construction,and a combination method based on lifting tree is proposed to realize feature non-linear combination construction.By using linear and non-linear combination methods,not only the feature construction automation is realized,but also the new features constructed can improve the training effect of the model better.Feature construction generates a series of linear and non-linear combinations of original features.After the basic scheme of feature construction is formed,the traditional parameter adjustment method is usually completed manually,which is time-consuming,laborious and lack of theoretical basis.A method of parameter automatic adjustment through Bayesian optimization learning is designed to further improve the efficiency of parameter adjustment and model training effect.The detailed design of software defect prediction system based on Automatic Feature Engineering is given.After training the software defect prediction model based on automatic feature engineering using the authoritative data set MDP in the field of software defect prediction,the prediction accuracy,accuracy and recall of the model are compared with those of the defect prediction model without automatic feature engineering.The results show that the former is superior to the latter in performance.