Stochastic-Aware Conformance Checking: An Entropy-Based Approach

Abstract

Business process management (BPM) aims to support changes and innovations in organizations' processes. Process mining complements BPM with methods, techniques, and tools that provide insights based on observed executions of business processes recorded in event logs of information systems. State-of-the-art discovery and conformance techniques completely ignore or only implicitly consider the information about the likelihood of processes, which is readily available in event logs, even though such stochastic information is necessary for simulation, prediction and recommendation in models. Furthermore, stochastic information can provide business analysts with further actionable insights on frequent and rare conformance issues. In this paper, we propose precision and recall conformance measures based on the notion of entropy of stochastic automata that are capable of quantifying, and thus differentiating, frequent and rare deviations between an event log and a process model. The feasibility of using the proposed precision and recall measures in industrial settings is demonstrated by an evaluation over several real-world datasets supported by our open-source implementation.

Stochastic Process Discovery by Weight Estimation

“…Petri nets and Generalized Stochastic Petri Nets are well-established formalisms for modelling processes and a number of good overviews exist [4,8]. We use notations from the process mining literature [3,21].…”

“…Fodina [10] Inductive Miner [19] Split Miner [7] discover w freq w lhpair w rhpair w pairscale w fork w align estimate RSD [25] tEMSC [20] Entropy Recall & Precision [21] entity count edge count duration measures…”

“…We use a minimum probability mass parameter setting of 0.8 for feasibility. (2) Entropy Precision and Recall [21] are stochastic conformance measures based on the entropy of equivalent automata constructed from a given log or model. (3) Petri net entity count (places and transitions) and ( 4) edge count are used as structural simplicity measures, ensuring that conformance quality has not been achieved by sacrificing model simplicity and comprehensibility.…”

“…Unlike enhancement techniques, estimators can potentially change control flows when producing a stochastic process model. Stochastic process models have a corresponding, emerging, set of stochastic process conformance measures [16,20,21]. Consequently, the algorithms and models presented here are evaluated, in Sect.…”

A Spectrum of Entropy-Based Precision and Recall Measurements Between Partially Matching Designed and Observed Processes

“…prec (k,k) / recall (k,k) log k = 0 k = 1 k = 2 k = 5 k = 10 k = 20 k = ∞ 1 0.147 / 1.000 0.194 / 1.000 0.241 / 1.000 0.386 / 1.000 0.547 / 1.000 0.650 / 1.000 0.709 / 1.000 2 0.918 / 0.797 0.981 / 0.856 0.990 / 0.918 0.959 / 0.997 0.946 / 1.000 0.961 / 1.000 0.960 / 1.000 3 0.903 / 1.000 0.950 / 1.000 0.955 / 1.000 0.974 / 1.000 0.980 / 1.000 0.980 / 1.000 0.980 / 1.000 4 0.575 / 0.824 0.679 / 0.952 0.763 / 0.988 0.936 / 1.000 0.973 / 1.000 -0.995 / 1.000 5 0.025 / 1.000 0.034 / 1.000 0.046 / 1.000 0.087 / 1.000 0.145 / 1.000 --6 0.016 / 0.991 0.023 / 0.979 0.031 / 0.877 0.072 / 0.830 0.791 / 1.000 --7 0.030 / 1.000 0.043 / 1.000 0.057 / 1.000 0.095 / 1.000 0.137 / 1.000 -0.393 / 1.000 8 0.027 / 1.000 0.037 / 1.000 0.048 / 1.000 0.090 / 1.000 0.135 / 1.000 --9 0.020 / 1.000 0.025 / 0.861 0.032 / 0.386 0.083 / 1.000 0.859 / 1.000 -checking approaches have been recently proposed [16,15,20]; these account for the relative likelihoods of traces described in models and recorded in logs. Finally, methods that combine quantitative and qualitative conformance checking techniques visualize the conformance diagnostics over the process model and are based on alignment, token replay, or footprint matrices, refer to [3], [22], and [2], respectively.…”