Interactive Measures for Mining Understandable State Machines from Embedded Software: Experiments and Case Studies

Abstract

State machines are a commonly used formalism for specifying the behavior of a software component, which could also be helpful for program comprehension. Therefore, it is desirable to extract state machine models from code and also from legacy models. The main drawback of fully-automatic state machine mining approaches is that the mined models are too detailed and not understandable. In our previous work [1], we presented different measures for the interaction with the state machine extraction process, such as selecting a subset of state variables, reducing the state variable range and providing additional user constraints. These measures aimed to reduce the complexity of the mined state machines to an understandable degree. In this article, which is an extended version of [1], we evaluate the approach through a case study with twelve professional developers from the automotive supplier company Bosch. The study shows that adding our interactive measures to the model mining process leads to understandable state machines, which can be very helpful in a rich set of use cases in addition to program comprehension, such as debugging, validation and verification. Furthermore, we conduct an experiment to evaluate the required computation time of the interactive measures against the fully-automatic mining. The experiment shows that the interactive approach can drastically reduce the computation time.