DC-SRGM: Deep Cross-Project Software Reliability Growth Model
DC-SRGM: Deep Cross-Project Software Reliability Growth Model
复制标题
DOI：
10.1109/issrew.2019.00044
复制
发表时间：
2019-10
期刊：
影响因子：
--
通讯作者：
Kyawt Kyawt San-Kyawt;H. Washizaki;Y. Fukazawa;Kiyoshi Honda;Masahiro Taga;Akira Matsuzaki
中科院分区：
文献类型：
--
作者：
Kyawt Kyawt San-Kyawt;H. Washizaki;Y. Fukazawa;Kiyoshi Honda;Masahiro Taga;Akira Matsuzaki
先前的研究表明，用于跨项目预测的软件可靠性增长模型（SRGM）对于正在进行的开发项目更为实用。已经根据各种因素提出了几种软件可靠性增长模型（SRGM），以衡量可靠性，并有助于指出释放前剩余缺陷的数量。软件行业希望预测错误的数量，并监控新的或正在进行的开发项目的项目状况。但是，在初始开发阶段的项目中，可用数据受到限制。在这种情况下，应用SRGM可能会错误地预测错误的未来数量。本文提出了一种使用先前项目的功能来预测正在进行的开发项目的错误数量的新SRGM方法。通过案例研究，我们通过K-均值聚类来确定目标项目的类似项目，并形成新的培训数据集。基于复发性神经网络的深度长期记忆模型建立在获得的预测模型的新数据集上。根据实验结果，拟议的深跨项目（DC）SRGM的预测比传统的SRGM和正在进行的项目的深入SRGM的性能更好。
Previous studies have suggested that software reliability growth models (SRGMs) for cross-project predictions are more practical for ongoing development projects. Several software reliability growth models (SRGMs) have been proposed based on various factors to measure the reliability and are helpful to indicate the number of remaining defects before release. Software industries want to predict the number of bugs and monitor the situation of projects for new or ongoing development projects. However, the available data is limited for projects in the initial development phases. In this situation, applying SRGMs may incorrectly predict the future number of bugs. This paper proposes a new SRGM method using the features of previous projects to predict the number of bugs for ongoing development projects. Through a case study, we identify similar projects for a target project by k-means clustering and form new training datasets. The Recurrent Neural Network based deep long short-term memory model is built over the obtained new dataset for prediction model. According to experiment results, the prediction by the proposed deep cross-project (DC) SRGM performs better than traditional SRGMs and deep SRGMs for ongoing projects.