Comparison of search strategies for feature location in software models

Abstract Search-based model-driven engineering is the application of search-based techniques to specific problems that are related to software engineering that is driven using software models. In this work, we make use of measures from the literature to report feature location problems in models (size and volume of the model and density, multiplicity, and dispersion of the feature being located) and a set of search strategies (random search, iterated local search, hill climbing, an evolutionary algorithm, and a hybrid between an evolutionary algorithm and hill climbing). The goal is to analyze of the impact of different values that are used to describe the feature location problems and the performance obtained by the different search strategies. We apply the search strategies to 1895 feature location problems that are obtained from 40 industrial software product lines. This work shows that: 1) the best results overall are obtained by a hybrid between evolutionary algorithm and hill climbing; 2) the size of the search space has the greatest impact on the results obtained by the search strategies; and 3) the impact of each of the measures is not the same in the five search strategies. This work highlights the use of the search strategy that produces the best results. In addition, we provide recommendations on when to use each search strategy.