Model Independent Error Bound Estimation for Conformance Checking Approximation

Abstract

Conformance checking techniques allow us to quantify the correspondence of a process's execution, captured in event data, w.r.t., a reference process model. In this context, alignments have proven to be useful for calculating conformance statistics. However, for extensive event data and complex process models, the computation time of alignments is considerably high, hampering their practical use. Simultaneously, it suffices to approximate either alignments or their corresponding conformance value(s) for many applications. Recent work has shown that using subsets of the process model behavior leads to accurate conformance approximations. The accuracy of such an approximation heavily depends on the selected subset of model behavior. Thus, in this paper, we show that we can derive a priori error bounds for conformance checking approximation based on arbitrary activity sequences, independently of the given process model. Such error bounds subsequently let us select the most relevant subset of process model behavior for the alignment approximation. Experiments confirm that conformance approximation accuracy improves when using the proposed error bound approximation to guide the selection of relevant subsets of process model behavior.

PN-BBN: A Petri Net-Based Bayesian Network for Anomalous Behavior Detection

“…Considering the dramatic increase in the size of the log data recorded by information systems, Sani et al considered that exact alignment is not always required, and they discarded the reference model and used the determined correct behavior to construct prefix trees to identify problematic activities by obtaining consistency values under the approximate method through edit distance [12]. To ensure the accuracy of this approximate alignment, Sani gave a method to quantify the maximum approximation error of an arbitrary sequence using edit distances that can guide the extraction of log subsets from logs to be extracted [13]. Some studies have directly used alignment techniques to analyze the variability of logs concerning the reference model.…”