Mutation testing from probabilistic finite state machines

Abstract

Mutation testing traditionally involves mutating a program in order to produce a set of mutants and using these mutants in order to either estimate the effectiveness of a test suite or to drive test generation. Recently, however, this approach has been applied to specifications such as those written as finite state machines. This paper extends mutation testing to finite state machine models in which transitions have associated probabilities. The paper describes several ways of mutating a probabilistic finite state machine (PFSM) and shows how test sequences that distinguish between a PFSM and its mutants can be generated. Testing then involves applying each test sequence multiple times, observing the resultant output sequences and using results from statistical sampling theory in order to compare the observed frequency of each output sequence with that expected.