Mining Test Cases: Optimization Possibilities

Abstract—System monitors need oracles to determine whether observed traces are acceptable. One method is to compare the observed traces to a formal model of the sys-tem. Unfortunately, such models are not always available — software may be developed without generating a formal model, or the implementation deviates from the original specification. In previous work, we have proposed a learning algorithm to construct a formal model of the software from its test cases, thereby providing a means to transform test cases for offline testing into an oracle for monitoring. In this paper, we refine our learning algorithm with a set of state-merging rules that help to exploit the test cases for additional information. We discuss our approach in detail and identify optimization areas. Using the additional information mined from the test cases, models can be learned from smaller test suites