Process mining allows organizations to transform the data recorded during the execution of their processes into meaningful insights. These insights can help to detect problems and to improve the processes. Various process mining solutions have been developed, both for industrial and academic purposes. However, most of these solutions do not support the creation and execution of analytics workflows. The KNIME Analytics Platform (KNIME in short) is an open-source workflow-based analytics platform that supports various techniques in the field of data science. KNIME is widely used in numerous industries across many countries. This paper presents the process mining extension for KNIME, which integrates many powerful process mining algorithms into KNIME. Using the process mining extension of KNIME, process mining can be combined with other types of data science techniques available in
KNIME. process mining, data science, workflow
Process mining helps to analyze and monitor processes based on the events recorded during
their execution. The goal of process mining is to extract information from these events to allow
organizations to detect problems in their processes and improve decision-making. The field of
process mining [
The growing interest in process mining led to the development of numerous process mining
tools. ProM [
KNIME [
Thanks to its ease of use and high scalability (distributed executors on the KNIME Server, big data, and cloud integration), KNIME software is used by hundreds of companies in numerous industries. PM4KNIME integrates process mining algorithms implemented in ProM into KNIME. This allows for creating analytics workflows that combine process mining with the other types of data science techniques available in KNIME in a scalable, user-friendly environment. Instruction on how to install PM4KNIME can be found under https://pm4knime.github.io/userDoc/guides/ installation.
In this section, we provide an overview of PM4KNIME. A screen recording corresponding to this overview is available under https://pm4knime.github.io/userDoc/guides/demo.
KNIME stores data in table-based objects called DataTables. Algorithms in KNIME are implemented as nodes. A node can have multiple input ports, output ports, views, and dialogs. The input ports should be connected to the input objects required for executing the underlying algorithm of the node. The dialogs are used to set the parameters of the algorithm. After the successful termination of an algorithm, the output objects can be accessed through the output ports. A workflow in KNIME is a directed graph connecting multiple nodes through their input and output ports.
PM4KNIME currently supports:1 • Importing and exporting diferent objects (e.g., Petri net). • Exploring event logs (e.g., dotted chart). • Converting objects (e.g., XES logs into DataTables). • Event data manipulation (e.g., filtering). • Process discovery (e.g., inductive miner).
• Conformance checking (e.g., alignment-based replay). 1See https://hub.knime.com/pm4knime/extensions/org.pm4knime.feature/latest for a complete overview of all available functionalities.
• JavaScript visualizations (e.g., for Petri nets).
Most implemented nodes work on DataTables. Internally, we wrap around the implementations of the underlying process mining techniques from the plugins available in ProM.
Figure 1 shows a typical workflow in the field of process mining. It contains nodes for importing data from a CSV file, preprocessing, process discovery, JavaScript visualizations of the discovered models, model and data transformation, and conformance checking. We applied this workflow to a real-life data set that records the execution of a ticketing management process [4]. Further workflow examples are available on the KNIME Hub under: https://kni.me/s/VJqKc-EypN7Jkrl2.
In [5], RapidProM was introduced as an extension of RapidMiner. It integrates process mining algorithms from ProM into the workflow-based platform RapidMiner. The idea of [ 5] is similar to our contribution, but PM4KNIME provides some features that diferentiate it from RapidMiner.
We adapted some process mining techniques not supported in RapidMiner (e.g., hybrid
Petri net miner). Moreover, most implemented algorithms in PM4KNIME work on DataTables
(not XES logs). We wrapped around the implementations of the underlying process mining
techniques in ProM. In data science, data is often stored in table-based files (e.g., CSV files) that
can be easily imported as DataTables in KNIME. Applying process mining algorithms directly
on DataTables improves the time performance because KNIME uses powerful caching strategies
that ensure high scalability when processing large DataTables [
The KNIME Server provides many valuable features for organizations, such as automated and distributed executions of workflows, deployment options, workflow management, and monitoring functionalities. PM4KNIME provides JavaScript visualizations for the diferent types of supported process models. This allows for building interactive web-based applications using the deployment options on the KNIME server.
Both RapidProM and PM4KNIME allow for saving workflows to be reused later. However, PM4KNIME additionally supports the serialization of intermediate results. Each node in a KNIME workflow processes its entire input data and permanently stores its output before forwarding it to the successor nodes. By saving a workflow, the settings of all nodes and all already generated (intermediate) objects are stored together with the workflow structure. Therefore, it is possible to stop the execution of a KNIME workflow at any node. The workflow can be modified and saved to be resumed later without needing to re-execute already executed nodes that are not afected by any modifications. For all implemented (intermediate) objects in PM4KNIME, we created internal importers and exporters to support the serialization of results.
In this paper, we introduced the process mining extension of KNIME (PM4KNIME). PM4KNIME integrates process mining algorithms that are implemented in the academic process mining tool ProM into a workflow-based data science analytics platform that is widely used in industry. The process mining extension of KNIME supports many techniques for process discovery, conformance checking, event data manipulation, and visualization of process models.
As future work, we aim at adapting further algorithms to work directly on DataTables instead of XES logs (e.g., conformance checking algorithms). Moreover, we aim at supporting more types of process models (e.g., BPMN models) and integrating more process mining algorithms from ProM and/or other academic tools like PM4Py (http://pm4py.org/).
The authors would like to thank Kefang Ding and Ralf Riesen for their contribution to PM4KNIME. and Knowledge Organization, Springer, 2007, pp. 319–326. URL: https://doi.org/10.1007/ 978-3-540-78246-9_38. doi:1 0 . 1 0 0 7 / 9 7 8 - 3 - 5 4 0 - 7 8 2 4 6 - 9 \ _ 3 8 . [4] M. Polato, Dataset belonging to the help desk log of an Italian Company (2017).
URL: https://data.4tu.nl/articles/dataset/Dataset_belonging_to_the_help_desk_log_of_an_ Italian_Company/12675977. doi:1 0 . 4 1 2 1 / u u i d : 0 c 6 0 e d f 1 - 6 f 8 3 - 4 e 7 5 - 9 3 6 7 - 4 c 6 3 b 3 e 9 d 5 b b . [5] R. Mans, W. M. P. van der Aalst, H. M. W. Verbeek, Supporting Process Mining Workflows with RapidProM, in: L. Limonad, B. Weber (Eds.), Proceedings of the BPM Demo Sessions 2014 Co-located with the 12th International Conference on Business Process Management (BPM 2014), Eindhoven, The Netherlands, September 10, 2014, volume 1295 of CEUR Workshop Proceedings, CEUR-WS.org, 2014, p. 56. URL: http://ceur-ws.org/Vol-1295/paper5.pdf.