=Paper=
{{Paper
|id=Vol-2973/paper_187
|storemode=property
|title=Forward-Looking Process Mining
|pdfUrl=https://ceur-ws.org/Vol-2973/paper_187.pdf
|volume=Vol-2973
|authors=Mahsa Pourbafrani
|dblpUrl=https://dblp.org/rec/conf/bpm/PourbafraniA21
}}
==Forward-Looking Process Mining==
https://ceur-ws.org/Vol-2973/paper_187.pdf
Forward-Looking Process Mining
Mahsa Pourbafrani
Chair of Process and Data Science, RWTH Aachen University, Aachen, Germany
Abstract
Most process mining techniques are backward-looking. Based on historic data, they produce descriptive
process models and reveal performance and compliance problems. Forward-looking process mining fo-
cuses on turning the results of backward-looking techniques into prediction and actions. The current
techniques use detailed event data to provide insights. However, different angles of event data enable
capturing process behavior and underlying relationships between process variables for further future
analyses, e.g., daily arrival rate and resource efficiency. In this project, we aim to provide a forward-
looking framework for business processes to replay their processes and assess the effect of the actions
performed based on process mining insights. To do so, we use both detailed event data and aggregated
event data of processes over time, i.e., fine-grained event logs and coarse-grained process logs. Using
alternative simulation approaches such as System Dynamics (SD) it is possible to incorporate external
factors into the model using more coarse-grained logs. Furthermore, we focus on connecting the ef-
fects of strategic decisions and detailed business processes by providing a comprehensive simulation
framework, i.e., hybrid simulations.
Keywords
process mining, fine-grained event logs, coarse-grained process logs, scenario-based simulation, system
dynamics
1. Introduction and Problem Definition
Historical data on executions of business processes may be used to support the business owners
to analyze their processes. These event data have a wealth of information about the processes,
where backward-looking process mining techniques help to uncover these insights [1]. Forward-
looking process mining supports process owners in taking actions based on the provided
knowledge [2]. Forward-looking process mining is generally categorized in two categories:
prediction models using machine learning techniques, such as [3], and simulation techniques.
For instance, process mining capability to describe a process is used to enrich simulation models
and make foreseeing their future possible [4]. The current forward-looking techniques for
assessing the future state of a process do not fully cover the following aspects: (1) The majority
of approaches are at a fine-grained level (detailed) and do not take into account the impact
of quality-based factors on the process. Discrete Event Simulations (DES), for instance, are
incapable of capturing the impact of resource training on the efficiency of processes. (2) Various
cause and effect relations are invisible in the fine-grained event logs. Aggregating the event data
Proceedings of the Demonstration & Resources Track, Best BPM Dissertation Award, and Doctoral Consortium at BPM
2021 co-located with the 19th International Conference on Business Process Management, BPM 2021, Rome, Italy,
September 6-10, 2021
" mahsa.bafrani@pads.rwth-aachen.de (M. Pourbafrani)
© 2021 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073 CEUR Workshop Proceedings (CEUR-WS.org)
Figure 1: The overview of the existing (gray) and the proposed approaches for process simulation. We
use process mining techniques to make simulations more data-driven. Moreover, we combine classical
discrete event simulation (fine-grained models) with coarse-grained simulation models.
can reveal the different states of processes, e.g., the effect of daily workload on the resources’
efficiency. (3) The interaction between coarse-grained and fine-grained simulations of business
processes is not provided, e.g., the effect of an advertisement after three months on the idle
time of process activities and the arrival rate. Dotted lines in Figure 1 represent conventional
paths for designing simulation models of processes.
Our goal is to define and generate coarse-grained process logs at different levels and from
multiple aspects, shown in Figure 1. In this context, standard process mining techniques and
aggregated process analyses are referred to as Fine-grained process mining and Coarse-grained
process diagnostics, respectively. The process aggregated state and its behavior at that level
directly affect every single instance in the process. Following this step, data-supported process
diagnostics, as well as fine-grained and coarse-grained simulation models are derived. As a
result, a set of simulation models capable of performing the prescribing role for processes
are generated. The generated simulation models can be designed and verified to reflect the
processes. For instance, DES is a form of fine-grained process simulations using fine-grained
event logs and SD is a type of coarse-grained process simulations using coarse-grained process
logs. System dynamics, which is an aggregated simulation technique, models a system using
variables that describe the system over time [5]. For processes, instead of simulating each event
including single case arrival and execution, we simulate the process using variables such as the
daily arrival rate, or daily average service time.
Contribution to BPM Research Providing what-if analyses is an important step in Business
Processes Management (BPM) [6]. Exploiting historical event data of organizations supports
business process simulation in BPM [7]. The simulation parameters are extracted from the
processes, and they make simulation models more realistic. However, the provided models
are always detailed and try to mimic the exact behavior of the business processes, i.e., at the
detailed events. The goal of this project is to pave the way for comprehensive future analyses
of business processes by providing diagnostics on top of the coarse-grained business process
historical data. These insights and diagnostics are used to form data-driven simulations with
two levels of granularity: fine-grained and coarse-grained (strategical what-if analysis) models.
2. Proposed Solution (Framework)
We propose an approach to transform fine-grained event logs into coarse-grained process
logs. The transformed logs enable further analyses such as discovering hidden relations and
supporting simulation models generation of business processes at aggregated levels. Figure 1
illustrates the designed steps from a fine-grained event log to generate ultimate results, i.e., SD
models (1), process diagnostics (2), DES models (3), and hybrid process simulations (4). The
purpose of the forward-looking analysis of business processes determines different paths inside
the proposed approach. Given the process diagnostics at different levels such as discovering the
effect of workload on the resources per day, the coarse-grained simulation models in the form
of SD can be extracted [8].
The prototype of the main steps of the project is implemented as a tool (PMSD) in [9]. Our
proposed solution includes 7 different yet connected frameworks. The output of each framework
is based on a specific purpose in forward-looking process mining. The designed/implemented
frameworks are as follows:
• Preprocessing (coarse-grained process log): a set of possible process variables are defined
based on the process aspects in fine-grained event logs [10] and these process variables
are calculated over specific steps of time, e.g., hourly, or daily. To design accurate simula-
tion/prediction models, the time window for extracting the process variables highly affects
the quality of the simulation/prediction models. In [11], we used Time Series Analysis
techniques such as ARIMA [12] for finding the best window of time.
• Coarse-grained simulation model generation: we used linear/nonlinear correlations
between process variables over time to discover existing relationships at higher levels
and design the system dynamics models [13].
• Process diagnostics (aggregated levels): coarse-grained process logs represent the process
at different steps of time using different aspects, e.g., the daily arrival rate of cases and the
average waiting time in the process. Techniques such as Granger Causality [14] and Curve
Fitting [15] are used to discover underlying causes and effects relations. Furthermore,
these aggregated insights are also used for context-aware predictive process mining [16].
• Simulation and validation: after the simulation, the accuracy of the simulation models
can be evaluated by comparing the results to the values of process variables at each time
step, e.g., [17] performs an evaluation for a car production line.
• Simulation model refinement: by adding external and qualitative factors to the validated
SD models, strategic analyses will be possible, e.g., effects of a new advertisement strategy.
• Fine-grained simulation model: [4] is the pioneering work that introduces data-driven
simulation in process mining. This module implemented in [18, 19, 20] for different forms
of presentation, e.g, process trees or CPN models. This module automatically discovers
the process activity flow and enriches that with resources, capacities, and time aspects of
the process. These models using multiple techniques such as Earth-Mover’s distance [21]
and Performance Spectrum [22] are validated in [23], and can be used for regenerating
processes.
• Hybrid simulation of processes (SD and DES): the connection between two types of
simulation models enables applying the effect of high-level what-if analyses on the
fine-grained simulation of business processes.
3. Current Status and Challenges
The green steps in Figure 1 have been designed and evaluated in the forward-looking process
mining project. Designing and implementing the framework for process diagnostics at higher
levels of aggregation is the next step. We do so by exploiting customized methods to identify
the relationships between process variables on coarse-grained process logs, such as Vector
Autoregressive. The project’s second focus is on the automatic discovery of mathematical
equations for the SD models using statistical and machine learning methods, and the final step
is to implement hybrid simulations of processes, which are highlighted in yellow in Figure 1.
The current challenges to be addressed are mainly in providing use cases, determining
underlying equations for all the process variables in the SD models, and connecting two types
of simulations (red steps in Figure 1). There is always a trade-off between adding external
factors and the accuracy of simulation results. Therefore, real-world case studies with known
changes and effects, as well as process domain knowledge in the form of external variables
are required. Since these external variables are not quantifiable, the validity of the designed
simulation model cannot be easily assessed. Furthermore, using only the generated process
variables in the SD-Logs will limit scenarios and the main purpose of system dynamics modeling.
As a result, strategic decision-making involving quality-based variables is not entirely possible.
When connecting fine-grained simulations (DES) and coarse-grained simulations (SD) in the
hybrid simulation step, the following questions should be addressed. How can both simulation
models be synchronized? How should interaction points be defined and discovered in practice?
Which DES parameters, for example, are updated as a result of SD simulation? How should the
execution of two models be handled in practice? For instance, for generating CPN models and
updating them while simulating SD models at the same time. Furthermore, the user interaction
and designing the scenarios are the project’s open challenges.
Acknowledgments
Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under
Germany’s Excellence Strategy- EXC 2023 Internet of Production- Project ID: 390621612. We
also thank the Alexander von Humboldt (AvH) Stiftung for supporting our research.
References
[1] W. M. P. van der Aalst, Process mining - Data science in action, second edition, Springer
(2016). doi:10.1007/978-3-662-49851-4.
[2] W. M. P. van der Aalst, Process mining and simulation: A match made in heaven!, in:
Proceedings of the 50th Computer Simulation Conference, SummerSim 2018, 2018, pp.
4:1–4:12.
[3] N. Tax, I. Verenich, M. La Rosa, M. Dumas, Predictive business process monitoring with lstm
neural networks, in: Advanced Information Systems Engineering, Springer International
Publishing, Cham, 2017, pp. 477–492.
[4] A. Rozinat, R. S. Mans, M. Song, W. M. P. van der Aalst, Discovering simulation models,
Inf. Syst. 34 (2009) 305–327. doi:10.1016/j.is.2008.09.002.
[5] J. D. Sterman, Business dynamics: Systems thinking and modeling for a complex world,
McGraw-Hill (2000).
[6] K. Tumay, Business process simulation, in: Proceedings Winter Simulation Conference,
1996, pp. 93–98. doi:10.1109/WSC.1996.873265.
[7] M. Camargo, M. Dumas, O. González, Automated discovery of business process simulation
models from event logs, Decis. Support Syst. 134 (2020) 113284. doi:10.1016/j.dss.
2020.113284.
[8] M. Pourbafrani, S. J. van Zelst, W. M. P. van der Aalst, Scenario-based prediction of business
processes using system dynamics, in: On the Move to Meaningful Internet Systems:
COOPIS 2019 Conferences, 2019, pp. 422–439. doi:10.1007/978-3-030-33246-4\_27.
[9] M. Pourbafrani, W. M. P. van der Aalst, PMSD: Data-driven simulation using system
dynamics and process mining, in: Proceedings of Demonstration at the 18th International
Conference on Business Process Management, 2020, pp. 77–81. URL: http://ceur-ws.org/
Vol-2673/paperDR03.pdf.
[10] M. Pourbafrani, W. M. P. van der Aalst, Extracting process features from event logs
to learn coarse-grained simulation models, in: Advanced Information Systems En-
gineering - 33rd International Conference, CAiSE 2021, Melbourne, VIC, Australia,
June 28 - July 2, 2021, Proceedings, volume 12751 of Lecture Notes in Computer Sci-
ence, Springer, 2021, pp. 125–140. URL: https://doi.org/10.1007/978-3-030-79382-1_8.
doi:10.1007/978-3-030-79382-1\_8.
[11] M. Pourbafrani, S. J. van Zelst, W. M. P. van der Aalst, Semi-automated time-granularity
detection for data-driven simulation using process mining and system dynamics, in:
Conceptual Modeling - 39th International Conference, ER 2020, Proceedings, 2020, pp.
77–91. doi:10.1007/978-3-030-62522-1\_6.
[12] G. Box, G. M. Jenkins, Time Series Analysis: Forecasting and Control, Holden-Day, 1976.
[13] M. Pourbafrani, S. J. van Zelst, W. M. P. van der Aalst, Supporting automatic system
dynamics model generation for simulation in the context of process mining, in: Business
Information Systems - 23rd International Conference, 2020, pp. 249–263. doi:10.1007/
978-3-030-53337-3\_19.
[14] C. W. Granger, Investigating causal relations by econometric models and cross-spectral
methods, Econometrica: journal of the Econometric Society (1969) 424–438.
[15] A. Zielesny, From Curve Fitting to Machine Learning, volume 18, Springer, 2011.
[16] M. Pourbafrani, S. Kar, S. Kaiser, W. M. P. van der Aalst, Remaining time prediction
for processes with inter-case dynamics, in: 2nd International Workshop on Leveraging
Machine Learning in Process Mining ICPM 2021, Proceedings, 2021.
[17] M. Pourbafrani, S. J. van Zelst, W. M. P. van der Aalst, Supporting decisions in production
line processes by combining process mining and system dynamics, in: Intelligent Human
Systems Integration 2020, 2020, pp. 461–467. doi:10.1007/978-3-030-39512-4\_72.
[18] M. Pourbafrani, S. Jiao, W. M. P. van der Aalst, SIMPT: Process improvement using
interactive simulation of time-aware process trees, in: Research Challenges in Informa-
tion Science, Springer International Publishing, Cham, 2021, pp. 588–594. doi:10.1007/
978-3-030-75018-3_40.
[19] M. Pourbafrani, S. Vasudevan, F. Zafar, Y. Xingran, R. Singh, W. M. P. van der Aalst, A
python extension to simulate petri nets in process mining, CoRR abs/2102.08774 (2021).
URL: https://arxiv.org/abs/2102.08774. arXiv:2102.08774.
[20] M. Pourbafrani, S. Balyan, M. Ahmed, S. Chugh, W. M. P. van der Aalst, GenCPN: Automatic
generation of CPN models for processes (2021).
[21] S. J. J. Leemans, A. F. Syring, W. M. P. van der Aalst, Earth movers’ stochastic conformance
checking, in: Business Process Management Forum, Springer International Publishing,
Cham, 2019, pp. 127–143.
[22] V. Denisov, D. Fahland, W. M. P. van der Aalst, Unbiased, fine-grained description of
processes performance from event data, in: Business Process Management, Springer
International Publishing, Cham, 2018, pp. 139–157.
[23] M. Pourbafrani, W. M. P. van der Aalst, Interactive process improvement using enriched
process trees, 2021.