Leveraging a combination of machine learning and formal concept analysis to locate the implementation of features in software variants

Context: Recently, software variants are adopted to build software product lines in the industry. In this adoption, the available assets (features, source code, design documents, etc.) are reused to build a software product line rather than building it from scratch. The feature location is the first step in this adoption process. In the literature, numerous approaches were proposed to locate the implementations of features in the source code. Objective: However, these approaches are guided using feature-specific information, which is not always available, especially in legacy applications. In this study, a feature location approach is proposed without predefined feature-specific information. Method: The proposed approach incorporates a mathematical research technique called formal concept analysis with other proposed algorithms. This combination is empirically evaluated using a benchmark case study. Results: The obtained results demonstrate that this combination achieves promising results in terms of well-known used metrics in this area: Recall, Precision, and F-measure.Conclusion: Also, the results show that the approach effectively finds features implementation across software variants.