When following an extractive approach to build a model-based Software Product Line (SPL) from a set of existing products, features have to be located across the product models. The approaches that produce best results combine model comparisons with the knowledge from the domain experts to locate the features. However, when the domain expert fails to provide accurate information, the semi-automated approach faces challenges. To cope with this issue we propose a genetic algorithm to feature location in model-based SPLs. We have an oracle from an industrial environment that makes it possible to evaluate the results of the approaches. As a result, the proposed approach is able to provide solutions upon inaccurate information on part of the domain expert while the compared approach fails to provide a solution when the information provided by the domain expert is not accurate enough.
展开▼
