Analyzing software quality with limited fault-proneness defect data

Assuring whether the desired software quality and reliability is met for a project is as important as delivering it within scheduled budget and time. This is especially vital for high-assurance software systems where software failures can have severe consequences. To achieve the desired software quality, practitioners utilize software quality models to identify high-risk program modules: e.g., software quality classification models are built using training data consisting of software measurements and fault-proneness data from previous development experiences similar to the project currently under-development. However, various practical issues can limit availability of fault-proneness data for all modules in the training data, leading to the data consisting of many modules with no fault-proneness data, i.e., unlabeled data. To address this problem, we propose a novel semi-supervised clustering scheme for software quality analysis with limited fault-proneness data. It is a constraint-based semi-supervised clustering scheme based on the k-means algorithm. The proposed approach is investigated with software measurement data of two NASA software projects, JM1 and KC2. Empirical results validate the promise of our semi-supervised clustering technique for software quality modeling and analysis in the presence of limited defect data. Additionally, the approach provides some valuable insight into the characteristics of certain program modules that remain unlabeled subsequent to our semi-supervised clustering analysis.