Integration Testing of Components Guided by Incremental State Machine Learning

The design of complex systems, e.g., telecom services, is nowadays usually based on the integration of components (COTS), loosely coupled in distributed architectures. When components come from third party sources, their internal structure is usually unknown and the documentation is insufficient. Therefore, the system integrator faces the problem of providing a required system assembling COTS whose behaviour is barely specified and for which no model is usually available. In this paper, we address the problem of integration testing of COTS. It combines test generation techniques with machine learning algorithms. State-based models of components are built from observed behaviours. The models are alternatively used to generate tests and extended to take into account observed behaviour. This process is iterated until a satisfactory level of confidence in testing is achieved