Research About Software Defect Prioirty Prediction Model Based On AdaBoost-SVM Algorithm

Bug reports are submitted by developers and users during development and maintenance of the software system. It is a challenging task undoubtedly that managing the incoming deluge of new bug reports received in bug library of a large open source project. Traditionally, the method of forecasting bug priority is artificial discrimination, which not only consumes time and resources, but also results in the delay of the important bug resolution. In this paper, a kind of machine learning method is proposed to develop bug priority classifier, which could assign a proper priority to new bugs automatically, which is based on data mining classification technology.Firstly, this thesis explores the algorithm of text selection and the text feature formalization. It is because that the text feature of the software defect report include many important informations about software defects, which play an important role in text classification. Consequently, according to the traditional text selection algorithm, an improved text selection algorithm is put forward, which is based on semantic analysis. Secondly, Software defect priority is predicted based on support vector machine (SVM) algorithm. SVM could establish nonlinear classification model according to the training data sets, which could predict the classification of a new instance. SVM maps the original data which is linear inseparable into a higher dimension space, in order to find a hyperplane in the new mapping data sets. Finally, An improved algorithm is proposed in this paper, which is based on AdaBoost-SVM algorithm. SVM is taken as the weak learning machine of AdaBoost, which takes the advantage of AdaBoost algorithm, emphasizing those wrong samples. By this method, a strong classifier is achieved at last.Experimental evaluation of these two classifiers using precision, recall-rate and F-value measures proves that the proposed automatical priority classification based on AdaBoost-SVM algorithm is more effective than SVM algorithm, especially the recall-rate.