Quality Prediction and Assessment for Product Lines

In recent years, software product lines have emerged as a promising approach to improve software development productivity in IT industry. In the product line approach, we identify both commonalities and variabilities in a domain, and build generic assets for an organization. Feature diagrams are often used to model common and variant product line requirements and can be considered part of the organizational assets. Despite their importance, quality attributes (or non-functional requirements, NFRs) such as performance and security have not been sufficiently addressed in product line development. A feature diagram alone does not tell us how to select a configuration of variants to achieve desired quality attributes of a product line member. There is a lack of an explicit model that can represent the impact of variants on quality attributes. In this paper, we propose a Bayesian Belief Network (BBN) based approach to quality prediction and assessment for a software product line. A BBN represents domain experts' knowledge and experiences accumulated from the development of similar projects. It helps us capture the impact of variants on quality attributes, and helps us predict and assess the quality of a product line member by performing quantitative analysis over it. For developing specific systems, members of a product line, we reuse the expertise captured by a BBN instead of working from scratch. We use examples from the Computer Aided Dispatch (CAD) product line project to illustrate our approach.