The purpose of process mining is to discover a process model starting from a set of execution traces. Many algorithms have been developed in the last years to discover procedural models such as Petri Nets or EPCs; mining of declarative constraint-based models is instead a very novel topic. Mining in CLIMB does not aim to discover a complete process model, but to find a set of business rules (in the form of CLIMB constraints) which correctly “explain” execution traces. Because constraints specify what is mandatory as well as what is forbidden, it is more suitable to consider both correct and wrong instances as input of the mining process (i.e. to split the analyzed execution traces in two sub-sets representing successful and failed executions). In this way, the discovered CLIMB model will contain a set of constraints which correctly classify input traces.

Because SCIFF belongs to the Logic Programming setting, it is possible to exploit Inductive Logic Programming techniques to tackle this “declarative mining” problem. In particular, the algorithm ICL (Inductive Constraint Logic) has been adapted to mine SCIFF theories. The search space consists of rules that can be mined, and is delimited by a so called “language bias”. If the bias is tuned to cover exactly the different CLIMB constraints, then the mined theory can be rendered in a graphical way as an extended DecSerFlow model.