Empirically Evaluating Improved Heuristics for Test Data Generation from OCL Constraints using Search Algorithms

Efficiently generating test data is one of the key testing requirements of automated model-based test case generation. Keeping this in mind and driven by the needs of our industrial partners, we propose an improvement in heuristics that we previously defined to generate test data from OCL constraints using search algorithms. We evaluate our improved heuristics using two empirical studies with three search algorithms: Alternating Variable Method (AVM), (1+1) Evolutionary Algorithm (EA), and a Genetic Algorithm (GA). Furthermore, we used Random Search (RS) as a comparison baseline. The first empirical study was conducted using carefully designed artificial problems (constraints) to assess each individual heuristics. The second empirical study is based on an industrial case study provided by Cisco. The results of both empirical evaluations reveal that the effectiveness of the search algorithms, measured in terms of time to solve the OCL constraints to generate data, is significantly improved when using the novel heuristics presented in this paper. In particular, our experiments show that (1+1) EA with the novel heuristics has the highest success rate among all the analyzed algorithms, as it requires the least number of iterations to solve constraints to generate test data. Keywords-UML; OCL; Search-based testing; Test data; Empirical evaluation