Using Genetic Programming to Infer Output and Update Functions for EFSM Transitions

Extended finite state machines (EFSMs) provide a way to model systems with internal data variables. A key challenge here is inferring the functions which relate inputs, outputs, and internal variables, especially when such variables do not appear in the traces. In this investigation, we compare the accuracy of EFSMs inferred by our technique (available at https://github.com/jmafoster1/efsm-inference) with those inferred by MINT, the current state of the art (https://github.com/neilwalkinshaw/mintframework). Our results show that our technique produces more accurate models than MINT where the two techniques are comparable and that, unlike MINT, our technique is still applicable even when the output of particular transitions depends on values not present in the original traces.