Computing trace alignment against declarative process models through planning

TL;DR: It is demonstrated here how state-of-the-art classical planning technologies can be used for trace alignment by presenting a suitable encoding and results are reported using a real log from a financial domain.

Abstract: Process mining techniques aim at extracting non-trivial knowledge from event traces, which record the concrete execution of business processes. Typically, traces are "dirty" and contain spurious events or miss relevant events. Trace alignment is the problem of cleaning such traces against a process specification. There has recently been a growing use of declarative process models, e.g., DECLARE (based on LTL over finite traces) to capture constraints on the allowed task flows. We demonstrate here how state-of-the-art classical planning technologies can be used for trace alignment by presenting a suitable encoding. We report experimental results using a real log from a financial domain.