Research On Software Defect Prediction Method Based On Cost Sensitive Learning Adacost

With the development of society,software has increasingly become an indispensable item in human life,bringing convenience to human life while also hiding huge security risks.oftware testing is a necessary means to ensure software reliability.However,for software with complex functions,testing all software modules will inevitably consume a lot of manpower and material resources.Software defect prediction technology can solve this problem well.Machine learning algorithms can be used to build software defect prediction models through software measurement information and defect markings to predict defects in newly developed software modules,improve testing efficiency,and save testing resources.In this paper,a two-stage combined feature selection method is proposed to solve the problems of redundant metrics and irrelevant to classification tasks in software defect prediction.In the first stage,the filtering feature selection method with fast processing efficiency is used to pre select features with a large threshold.In the second stage,the wrapped feature selection algorithm is used to search the optimal feature subset based on the evaluation results of candidate feature subset in the verification set.Aiming at the class imbalance problem in software defect prediction,this paper proposes a software defect prediction model based on cost sensitive learning adacost.The model uses the cost sensitive learning mechanism of adacost algorithm to set different weight updating strategies for different categories of samples.For the defective samples with higher classification cost,the weight increases quickly and decreases slowly in the training process,so as to alleviate the impact of class imbalance on model training.Finally,the proposed method is tested on nine open source projects of NASA dataset,and the results show that the proposed method can effectively screen out feature subsets with high classification value,solve the class imbalance problem,and improve the classification performance of machine learning algorithm in software defect prediction.