Fuzzy adaptive tuning of a particle swarm optimization algorithm for variable-strength combinatorial test suite generation

Abstract

Combinatorial interaction testing is an important software testing technique that has seen lots of recent interest. It can reduce the number of test cases needed by considering interactions between combinations of input parameters. Empirical evidence shows that it effectively detects faults, in particular, for highly configurable software systems. In real-world software testing, the input variables may vary in how strongly they interact; variable strength combinatorial interaction testing (VS-CIT) can exploit this for higher effectiveness. The generation of variable strength test suites is a non-deterministic polynomial-time (NP) hard computational problem [1]. Research has shown that stochastic population-based algorithms such as particle swarm optimization (PSO) can be efficient compared to alternatives for VS-CIT problems. Nevertheless, they require detailed control for the exploitation and exploration trade-off to avoid premature convergence (i.e. being trapped in local optima) as well as to enhance the solution diversity. Here, we present a new variant of PSO based on Mamdani fuzzy inference system [2,3,4], to permit adaptive selection of its global and local search operations. We detail the design of this combined algorithm and evaluate it through experiments on multiple synthetic and benchmark problems.We conclude that fuzzy adaptive selection of global and local search operations is, at least, feasible as it performs only second-best to a discrete variant of PSO, called DPSO. Concerning obtaining the best mean test suite size, the fuzzy adaptation even outperforms DPSO occasionally. We discuss the reasons behind this performance and outline relevant areas of future work.

An Experimental Study of a Fuzzy Adaptive Emperor Penguin Optimizer for Global Optimization Problem

“…6, there are one pair, one pair, and three pairs for MC/DC coverage for variables A, B, and C, respectively. Three alternative MC/DC solutions (i.e., [0,1,3,5,7], [2,3,5,7], and [3,4,5,7]) are derived by combining the pairs except the repetition. The result is that any MC/DC findings may be transformed to appropriate test cases for all the predicates.…”

“…Moreover, they work with a randomly generated set of candidate solutions known as population and can be implemented easily due to their simple structure. collective behavior of the species in nature [4]. Without any centralized control, these species have the ability to position themselves for achieving their goal in a decent way [5].…”

Combinatorial Testing Approaches: A Systematic Review

“…Varying the intensity of the covering arrays can be used to detect these various interactions efficiently. This can provide various covering strengths to distinct groupings of variables, and thus can provide a realistic way to evaluate real applications [13] , [14].…”