Research And Optimization Of Software Fault Prediction Model Based On Machine Learning Method

As the scale and complexity of software increases,so does the cost of software testing.Therefore,effective prediction of software faults,so that software modules that may contain faults are prioritized to test,thereby reducing test cost overhead,has great practical significance.In this context,based on the two major goals of software fault prediction,namely,the probability of software modules being defective,and the possible ordering of defective software modules,the paper studies the software module fault classification prediction and software module fault ranking prediction..The specific work is as follows:Most of the software modules in the software fault data set are fault-free software modules,so the fault data set is a classified unbalanced data set.In order to solve the problem of classification imbalance,the paper studies the existing unbalanced processing methods and deeply analyzes the core processes of SMOTE method,ADASYN method,EasyEnsemble method and BalanceCascade.Finally,the paper has carried out sufficient experiments on a variety of different data sets and found that the EasyEnsemble sampling method works best.Software fault classification predictions fall into two categories,namely,predictions are defective and predictions are non-defective.In the paper,the paper firstly studied the current popular integrated learning methods,including Bagging and Boosting methods.Then,it studied a variety of Bagging-based integrated learning methods and a variety of Boosting-based integrated learning methods.Subsequently,the thesis The Stacking strategy is used to integrate a variety of integrated learning methods to construct a new model.Finally,the paper designs sufficient experiments for a variety of different data sets.The experimental results show that the integrated learning model based on Stacking can effectively improve the model classification prediction ability.The paper predicts the number of faults in the software module through the model,and then sorts.