基于改进随机森林的软件故障预测模型研究-计算机软件与理论专业论文.docxVIP

摘要 摘要 摘要 在软件开发过程中，准确地预测有故障倾向的软件模块可以帮助测试人员以 及开发人员高效地发现软件故障。对于大型系统，软件故障预测就更加重要，因 为它能够帮助开发人员在开发过程中将精力和资源集中使用在容易出现故障的模 块，从而更合理地分配资源。 目前软件故障预测模型主要是通过机器学习算法来构建的，软件故障预测主 要是根据软件质量属性以及软件的历史故障数据来对被测系统的软件模块进行故 障预测。本文首先对软件质量属性进行了研究，包括方法级属性集合 McCabe 和 Halstead，面向对象(OO)属性集合。本文提出的软件故障预测模型采用方法集属性 与面向对象属性相结合的方法对软件进行度量。针对软件故障数据的不平衡性问 题，本文提出了一种改进的随机森林算法，在对模型进行训练前首先对训练数据 集进行平衡化处理，以减少训练数据对模型准确度的影响，模型生成后在保证预 测准确率较高的情况下精简模型，提高预测效率。应用本文提出的算法对 Eclipse 中的数据进行处理，实验表明本文提出的改进算法性能高于其他现存的机器学习 算法。 关键词：软件质量 软件故障预测模型 随机森林算法 决策树 Abstract Abstract Abstract During the development process of software, accurate prediction of fault prone modules enables the testers and developers to identify the defects efficiently. Software defect prediction models are more valuable for the large-scale system, where the developers need to focus their attention and resources on the fault prone modules in the system under development. Currently, many machine learning algorithms have been developed and applied for software defect prediction. Software prediction models mainly use software quality attributes and the historical defect records to discover the fault prone models in the system under testing. Firstly in this thesis, software quality metrics, including method level metric sets (McCabe, Halstead) and the Object-Oriented metric sets are studied. The software prediction model proposed in this thesis uses the combination of method level metric sets and object-orient metric sets to measure software quality. To solve the problem of imbalanced data, this thesis presents an improved random forest algorithm, which will make the training data balance before building the prediction model. This can reduce prediction bias caused by the imbalance data. The improved random forest also simplifies the model in order to obtain high efficiency of handling big data sets. Using Eclipse defect data as testing set, the improved random forest algorithm has better performance than other machine learning algorithms. Keywords: Software Quali