Improving trace accuracy through data-driven configuration and composition of tracing features

Software traceability is a sought-after, yet often elusive quality in large software-intensive systems primarily because the cost and effort of tracing can be overwhelming. State-of-the art solutions address this problem through utilizing trace retrieval techniques to automate the process of creating and maintaining trace links. However, there is no simple one- size-fits all solution to trace retrieval. As this paper will show, finding the right combination of tracing techniques can lead to significant improvements in the quality of generated links. We present a novel approach to trace retrieval in which the underlying infrastructure is configured at runtime to optimize trace quality. We utilize a machine-learning approach to search for the best configuration given an initial training set of validated trace links, a set of available tracing techniques specified in a feature model, and an architecture capable of instantiating all valid configurations of features. We evaluate our approach through a series of experiments using project data from the transportation, healthcare, and space exploration domains, and discuss its implementation in an industrial environment. Finally, we show how our approach can create a robust baseline against which new tracing techniques can be evaluated.