Applying Many-objective Algorithms to the Variability Test of Software Product Lines

• Thiago do Nascimento Ferreira University of Michigan
• Silvia Regina Vergilio UFPR
• Marouane Kessentini University of Michigan

Resumo

The problem known as Variability Test of Software Product Line (VTSPL) is related to the selection of the most representative products for the SPL testing. This is an optimization problem because a possible exponential number of products can be derived from the SPL variability model, such as the Feature Model (FM). In the literature many works are dedicated to this research subject, each one applying a different search-based algorithm and using distinct criteria. However, there is no study encompassing all these criteria at the same time. To this end, this paper investigates the use of two Many-Objective Evolutionary Algorithms (MaOEAs). We apply the algorithm NSGA-III, widely used for many-objective algorithms, and the algorithm PCA-NSGA-II, a reduction dimensionality algorithm, which uses the Principal-Component Analysis (PCA) in combination with NSGA-II, to evaluate the objectives used in the literature for the VTSPL problem. PCA-NSGA-II reduces the search space dimensionality by eliminating the redundant objectives. The analysis shows the importance of some objectives such as the number of alive mutants, similarity between products, and unselected features. NSGA-III reaches the best results regarding the quality indicators for all instances, but taking a longer time. Besides, PCA-NSGA-II can find different solutions in the search space that are not found by NSGA-III.

Palavras-chave: many-objective problems, dimensionality reduction, Software product line testing