Research On Incremental Model Method Of Software Defect Prediction

With the rapid development of IT industry,software has become a necessity in people's life.Application software that can meet users' work,life and entertainment has been developed one after another.Users also put forward higher requirements for software quality.As an important factor affecting software reliability,software defects have become a problem that must be considered in software engineering.Software defect prediction technology can accurately find software modules with potential defects.As an important way to improve software reliability and software testing efficiency,it has gradually become a research hotspot in the field of software engineering.The traditional static batch learning software defect prediction method uses the historical software data warehouse for one-time modeling,and then predicts the defects of unpublished software modules in the process of software development.Nowadays,software data is generated in the form of dynamic data flow.Traditional static batch learning needs to retrain the model many times when facing software data flow,which is very inefficient and causes a waste of resources.Therefore,incremental learning provides a new idea for software defect prediction technology of dynamic streaming software data.The software defect prediction method based on incremental learning can process dynamic software data in real time,classify efficiently and accurately,and make up for the shortcomings of traditional static batch learning in front of dynamic software data flow.The main contributions of this paper are:(1)Aiming at the randomness and imbalance of software data flow,a differential software data window extraction algorithm is proposed.Firstly,the algorithm intercepts the real-time software data stream based on time series;Secondly,the feature vector of data is constructed,and the difference between data is used to update the data window;Finally,considering that there is still class imbalance in the differential software data window,the under sampling technology is used to alleviate the imbalance between classes.The simulation experiment is carried out using NASA software defect prediction open data set.The experimental results show that the differential software data window extraction algorithm improves the performance of incremental model classification.(2)Considering that the incremental model increases with the increase of incremental times in integrated incremental learning,which finally leads to the problems of difficult system load and low classification efficiency,a software defect prediction incremental model-based classifier weight learning algorithm is proposed.With the increment process,the algorithm first evaluates the k-statistic coefficients of each base classification and other base classifiers and the classification accuracy of each base classification;Secondly,k-statistic coefficient and accuracy are combined to determine the weight of base classifier;Finally,when the weight of the base classifier is 0,the base classifier is deleted to control the scale of the algorithm.The simulation experiment is carried out using the NASA software defect prediction open data set.The experimental results show that the software defect prediction incremental model-based classifier weight learning method is significantly better than the defect prediction of in project defect prediction and traditional incremental learning in PD index.Finally,the work of this paper is summarized,and the future research work is prospected.