Conformance checking techniques compare process models with real execution data to assess their alignment. Alignments are valuable for calculating conformance statistics, but exact solutions can be computationally expensive for large event data sets. This paper presents an easy-to-use plug-in for the ProM process mining framework that approximates alignment values. The plug-in takes an event log and process model as input, providing an approximate alignment value along with bounds for the actual alignment. Diagnostic information on problematic activities is also provided. Three approaches are ofered: subset selection, simulation, and log-to-log comparison. This plug-in enables eficient conformance assessment, overcoming computational challenges for large event data sets.
Conformance checking, a fundamental aspect of process mining, focuses on assessing the adherence of
a designed or discovered process model to real process executions [
In numerous applications, the computation of alignment values is required multiple times. For example, when seeking an appropriate process model for an event log, various process discovery algorithms with diferent settings are employed to discover multiple process models. The alignment techniques are then used to assess the fit between each process model and the event log. However, traditional alignment methods tend to be time-consuming, especially when dealing with large event data sets. This limitation makes it impractical to analyze numerous candidate process models within a limited timeframe. Therefore, by reducing the computation time of alignments, a greater number of candidate models can be considered for evaluation. Additionally, in many cases, obtaining precise alignment values is unnecessary, and having a quick approximation or a close lower/upper bound would sufice.
Recently, various approaches for alignment approximation have been proposed [
This demo paper presents an easy-to-use Alignment Approximator plug-in that utilizes three current approaches to approximate alignments, ensuring flexibility and accuracy. By ofering bounds for the actual alignment, the plug-in provides users with valuable insights into the reliability of the approximated value. Additionally, users have the option to consider an event log as some possible behavior of the process model that can be achieved by simulation, enabling alignment computations for process models represented in various notations, including Petri nets. It should be noted that it is not required that the event log that represents the model contains all model’s behavior as we approximate the alignment cost.
The proposed plug-in encompasses three distinct approaches for approximating alignments: • Subset Selection: This approach allows users to select a subset of model behavior based on customizable parameters, efectively approximating the alignment value with by computing alignment of limited number of variants in the event log. • Simulation: Leveraging process model simulation, this approach estimates the alignment value by configuring simulation parameters to achieve optimal results. This method does not need to compute any alignment. • Log-to-Log Comparison: By comparing the behavior captured in two event logs, this approach approximates the alignment value. It is particularly useful when evaluating alignment without the requirement of discovering a process model, and when reliable variants of traces are available.
These three approaches enable users to approximate alignment values eficiently while also providing diagnostic information.
The remainder of this paper is structured as follows. In Section 2, we discuss the approximation methods at an abstract level. Moreover, Section 3 explains how to use the developed tool. Furthermore, Section 4 briefly describes the maturity of Alignment Approximator. Finally, Section 5 concludes the paper presents some directions for extending the implementation.
The general idea of the used sampling method is presented in Fig. 1. The general idea is instead of
using the whole process model that is a set of sequences over a set of activities (i.e., {}⊆ℳ⊆ * ),
we proposed to use a subset of process model (i.e., {}⊆ℳ ′⊆ℳ ). As it shows in [
To generate the process model subset, we consider three approaches. In subset selection, we can
select some of the variants in the event log and compute their alignments using the classical alignment
method [
m1 〈 …, a, b, c, ...〉
m2 〈 …, a, b, b, ...〉
m3 〈 …, a, c, b, ...〉 . . .
mn 〈 …, c, b, c, ...〉
Process Model ( )
Model subset
( )
and similarity to other variants and selecting them randomly. The drawback of this method is we
should describe the given process model with the Petri net notation. This approach and how we can
compute bounds for the actual alignment cost are presented in [
The second approach to generate the subset of model traces is simulation. This simulation could be
done randomly or be guided by the behavior and their probabilities in the event log [
In the last approach, i.e., log-to-log comparison, we consider a process model as an event log that is a multiset over a set of sequences of activities. In other words, let denotes a set of activities, we define an event log as ∈ℬ(* ). In this regard, we first find the set of unique variants in the event log, i.e., . As in this approach, we directly use some possible traces of the process model, even if there is no reference process model and just some of the correct behaviors of the process (e.g., some of the valid variants) are known, the proposed method is able to approximate the conformance value. Using this approach, we are able to compute the alignment of process models with any notations until there is a subset of their simulation. It is also possible to give the process model behaviors as a CSV file and later convert it to an event log.
Using all three approaches, the plug-in returns an approximation for the alignment cost and upper and lower bounds for the actual alignment. Note, it returns these fitness values that are directly computed based on the alignment costs. Moreover, the plug-in provides some diagnostic information about the problematic activities and how many times each activity has sync/async moves.
We have developed the Alignment Aproximator tool as a plug-in in the ProM [
As mentioned earlier, there are three methods available for generating the subset of a process model. These methods involve: 1) aligning a process model with selected variants from the event log (subset selection), 2) simulating the behavior of the process model (simulation), and 3) considering a process model as a set of activity sequences (log-to-log comparison).
The provided plug-in receives two inputs that are event log and process model. Users can give a
(simulated) event log or a Petri net to have a process model. A snapshot of this plug-in and its inputs
are presented in Fig. 2 If the user provides a Petri net for the input process model, the plug-in provides
two possibilities to the user to approximate the alignment value, i.e., simulation and subset selection.
If the user prefers the subset selection method, he/she can select diferent selection strategies, e.g.,
clustering, frequency, and similarity. By selecting the simulation approach, users can adjust the
simulation by diferent methods, e.g., the number of simulated traces and the type of summarization.
For more information about how to adjust the settings, please refer to [
The output of this plug-in is an approximation of the alignment, i.e., the approximated fitness, upper and lower bounds for the actual fitness, and the number of synchronous and asynchronous moves for diferent activities. Two snapshots of the output of the Alignment Approximator tool are presented in Fig. 3a and Fig. 3b.
A video that describes how to use this tool is presented in https://youtu.be/eJBBuNhFmC4. Moreover, in https://github.com/fanisanim/AlignmentApproximator, we provide a comprehensive guide on how to efectively utilize the Alignment Approximator tool.
The proposed tool integrates multiple approximation algorithms, making it user-friendly and
accessible to end-users. These algorithms have been extensively applied to approximate event logs
from diverse real-world datasets, showcasing their ability to enhance the performance of alignment
computations [
This demo paper presents a ProM plug-in that approximates alignments and provides bounds for alignment costs, along with diagnostic information about synchronous and asynchronous moves in activities. Developed within the ProM platform, it ofers three diferent methods for alignment approximation: subset selection, simulation, and log-to-log comparison. The log-to-log comparison approach extends the tool’s applicability to process models with diferent notations by enabling simulation. Additionally, we provide a video and a tutorial that ofer step-by-step instructions on how to use the tool efectively. To advance this research, we aim to provide a method that ofers acceptable approximation error bounds and adjusts method settings based on event log characteristics. Moreover, we want to show the diagnostic information of the alignment on process models that helps analysts detect the process’s problematic part.
Funded under the Excellence Strategy of the Federal Government and the Länder. We also thank the Alexander von Humboldt (AvH) Stiftung for supporting our research.