期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2016
卷号：94
期号：2
出版社：Journal of Theoretical and Applied
摘要：To begin with, this research defines Software Quality Prediction System (SQPS) as a system composed of a Classification Algorithm (CA) and a Software Quality Measurement Model (SQMM). Machine Learning applications in software quality measurement are expanding as research intensifies in two directions, the first direction focuses on improving the performance of CAs while the other direction concentrates on improving SQMMs. Despite of the increasing attention in this area, some well-designed SQPSs showed considerable false predictions, which could be explained by faults in the design of the CA, the SQMM, or the SQPS as a whole. In this context, there is a debate on which CA is better for measuring software quality, as well as there is a debate on which SQMM to follow. To start with, the research studied an original dataset of 7311 software projects. Then, the research derived quality measurements from the ISO 9126 Quality Model and developed the SQMM accordingly. Next, the research compared statistical measures of performance of four CAs, using WEKA and SPSS. Our experimental results showed that ISO 9126 is general, but flexible enough to act as a SQMM. Despite of their convergent performance, our experiments showed that Multilayer Perceptron Network (MLPN) have more balanced predictions than Naive Bayes does. Following a rarely researched approach, the SQPS predicted five levels of software quality instead of making a binary prediction, limited with defect or non-defect software.
关键词：Software Quality Prediction System (SQPS); ISO 9126 Software Quality Model; Multilayer Perceptron Network (MLPN); Classification Algorithm (CA); Software Quality Measurement Model (SQMM); Machine Learning.
