【正文】 e used, we employ thedisagreementbased semisupervised learning paradigm[10] to exploit the unlabeled examples in U. In detail, after the initial ensemble of classi175。er. Such superiority is more obvious whenthe training set is small[37] and the class distribution isimbalanced[48]. Thus, by exploiting the generalizationpower, the trained ensemble from L is able to identifysome minorityclass examples from U e174。c choice of the ensemble learningparadigm is that an ensemble of classi175。rst requirement, we train multiple classi175。uence of overwhelming number of the newly labeledmajorityclass examples should be further reduced inorder to improve the sensitivity of the learner to theminority examples after its re175。nement, and this leads toeven poorer sensitivity to the minorityclass. As the iterative semisupervised learning proceeds, the learnedmodel would be biased to predict every example to themajorityclass.In order to successfully conduct iterative semisupervised learning on the imbalanced data, the learnershould have the following two properties. First, thelearner should have strong generalization ability, suchthat even if provided with a small labeled training setwith imbalanced class distribution, the learner wouldnot have zero sensitivity to the minorityclass examplesduring the automatically labeling process。 PD(yi = 161。 +1g is the class label. Conventionally, +1 denotes the minority class (., \defective in softwaredefect detection). Thereinafter, we refer to class +1 asthe minorityclass and 161。 xNg denote the set of unlabeled examples,where xi is a ddimensional feature vector, and yi 2f161。 ym0 )g denotethe set of labeled examples and let U = fxm0+1,xm0+2。 : : : 。 (x2。cial to software defectdetection.3 Proposed ApproachLet L = f(x1。 using ensembletechnique for exploratory undersampling to avoid theremoval of useful majority class examples[48].The classimbalance learning method is seldom usedin software defect detection. Recently, Pelayo andDick[9] studied the e174。 bining di174。ective to classimbalance problems. Sophisticated methods can be employed to balance theclass distribution, such as adding synthetic minorityclass examples generated from the interpolation ofneighboring minorityclass examples[43]。ers to the minority class may bevery low if directly learning from the imbalanced data.To achieve better sensitivity to the minority class, theclassimbalance problem should be explicitly tackled.Popular classimbalance learning techniques includesampling[11。cation, Zhou and Li[39] also adapted thedisagreementbased paradigm to semisupervised regression. Disagreementbased semisupervised learningparadigm has been widely applied to natural languageprocessing (., [40]), information retrieval (., [4142]), puteraided diagnosis (., [37]), etc.Few researches applied semisupervised learning tosoftware defect detection, where the labeled trainingexamples are limited while the unlabeled examples areabundant. Recently, Seliya and Khoshgoftaar[8] applied a generativemodelbased semisupervised learning method to software defect detection and achievedperformance improvement. Note that [8] adopted agenerative approach for exploiting unlabeled data whilethe proposed method adopts a discriminative approach.Thus, we did not include it in our empirical study forthe purpose of fair parison. Learning from Imbalanced DataIn many realworld applications such as softwaredefect detection, the class distribution of the data is imbalanced, that is, the examples from the minority classare (much) fewer than those from the other class. Sinceit is easy to achieve good performance by keeping themajorityclass examples being classi175。ers agree on its labeling. Later,Li and Zhou[37] further extended the idea in [38] bycollaborating more classi175。ers to exploit unlabeled data, where an unlabeledexample is labeled and used to teach one classi175。dently predicted unlabeled examples. Later, Goldman and Zhou[36] proposed an algorithm which doesnot require two views but require two di174。ers learned from two su177。ort and expertise arerequired. Semisupervised learning[2425] is a machinelearning technique where the learner automatically exploits the large amount of unlabeled data in additionto few labeled data to help improving the learning performance.Generally, semisupervised learning methods fallinto four major categories, ., generativemodelbased methods[2628], low density separation basedmethods[2931], graphbased methods[3234], anddisagreementbased methods[3539].Disagreementbased methods use multiple learnersand exploit the disagreements among the learners during the learning process. If majority learners are muchmore con175。cult to collect). It has beenshowed that more accurate detection could be achievedeven if only one of such characteristics is considered during learning[89]. Since these characteristics are usuallyintertwined with each other, better performance couldbe expected if carefully considering them together during learning, which is what we do in this paper.萬(wàn)方數(shù)據(jù)330 J. Comput. Sci. amp。cial neural networks[1。cation and regression trees[3。ow graph of the module[18].In the past decade, machine learning has beenwidely applied to construct predictive models basedon the extracted software metrics to detect defects inthe software modules. Typical methods include linear or logistic regression[7。rst category leveragethe execution information to identify suspicious program behaviors for defect detection[1214], while methods in the second category elaborate to extract staticcode properties, which are usually represented by a setof software metrics, for each module in the softwaresystem[7。ective for software defect detection. Its performance is better thanboth the semisupervised learning method that ignoresthe classimbalance natur