This paper presents a comprehensive empirical evaluation of four canonical bioinspired algorithms– genetic algorithm (GA), particle swarm optimization (PSO), ant colony optimization (ACO), and differential evolution (DE)–applied to search-based software testing (SBST). The main objective is to analyze the sensitivity of key configuration parameters, particularly population size, and assess their influence on the number of evaluations required to achieve full code coverage. A full factorial experimental design was conducted on three artificial systems under test (SUTs), each with varying structural complexity. Results were statistically analyzed using ANOVA and visualized through boxplots to detect significant parameter effects and interactions. Findings reveal that GA and ACO exhibit notable sensitivity to population size, while PSO and DE demonstrate greater robustness, with DE showing SUT-dependent behavior. The structural complexity of the SUTs consistently influenced algorithm performance. This work contributes practical insights for tuning bioinspired algorithms in SBST and offers a replicable methodology for future sensitivity analyses in automated test generation.
Domínguez-Isidro et al. (Mon,) studied this question.