An Empirical study of Exploring Relevant Metrics to Assess Software Product Quality

This paper proposes an empirical study of exploring relevant and operable metrics to assess software product quality based on quality characteristics of ISO/IEC 25010. Both data-driven and Goal-Question-Metric approach are applied to excavate as many full-life-cycle metrics as possible. Goal-Question-Metric approach is applied to identify software metrics affecting product quality in a specific domain. Machine learning algorithm that support incremental training is applied to learn the relationship between code metrics and quality characteristic marks from historical data. Thus full-life-cycle software metrics are identified and we build a software quality assessment model based on historical code metric data.Two case studies are conducted based on actual projects from the past 2 years to verify the feasibility of methodology, including an empirical investigation of mostly concerned quality factors in an electronic equipment software institute, a collection of code metrics and quality characteristic marks of 82 aviation embedded software projects, as well as a comparation of the performance of logistic regression, k-Nearest Neighbor and BP neural network algorithm for quality assessment. Those cases are performed with actual data in actual institutes and projects. Additionally, a support tool for project data management and quality assessment is developed.Through feedback from cooperative engineers, we still see room for improvements to fill the gap between methodology and actual software engineering process. Nevertheless, the empirical validation shows the feasibility of the metrics for quality assessment.