Process Mining is a data mining technique that allows us to build process models from event logs. For example, we can use audit trail logs, transaction logs, or any other kind of event logs to build models that approximate the process behind the series of events. Process models give us a compact end-to-end representation of the processes at play that are 'causing' the observed logs. Process mining started in the business process domain, but it has since been used in various other domains such as health and education. For more info, visit the Process Mining website.
There are a few different tools available to do process modeling. Two of the most famous ones are ProM and Disco. ProM is free and open source, and contains a plethora of process mining algorithms. We figured out that out of all of the available algorithms, Fuzzy Process Mining was probably the most useful process modeling algorithm when it came to mining real world data as this algorithm provided many different abstractions over the process. Fuzzy Process Mining algorithm was not available in R, so we built fuzzymineR to use the Fuzzy Process Mining algorithm available in ProM.
In fuzzymineR, to build a process model from event log data, a user has to only provide a data.frame that contains an event log. fuzzymineR provides a way to create an object of class bupaR::eventlog (or you can create the object directly yourself) and that object can be used to create a Fuzzy Process Model. Once the model is mined, you can visualize the model by setting the parameters of the visualization function and get the desired level of abstraction that you would like to work with. For process discovery, fuzzy mining algorithm is considered one of the best algorithms because it can deal with less-structered event data and it allows you do look at the process from different abstraction levels. Data complexity is a typical issue and fuzzy mining algorithm often saves you from creating extremely complex spaghetti models.
Our package carries ProM JAR files within it. Right now, we are using ProM 6.10 JAR files.
Your system should have Java 8 or above. We tested the package with Java 8 and Java 15. For Windows, please ensure that you are using a 64-bit version of Java. If you encounter any errors in the mine_fuzzy_model function, make sure you have verbose = TRUE so that you can see the problem.
create_eventlog: Creates eventlog object from adata.framemine_fuzzy_model: Applies fuzzy mining algorithm on the eventlog object and returns a list of metricsviz_fuzzy_model: Creates process model after processing the list of metrics
To install the package, use the following command:
devtools::install_github("PlaypowerLabs/fuzzymineR")If the installation gives you an error saying sh: rm: command not found sh: mv: command not found then please check if your ~/.Renviron file makes any modifications to the $PATH variable. If it does, please remove them and try the installation again. Here are two links that are related to this error:
https://community.rstudio.com/t/sh-rm-command-not-found-sh-mv-command-not-found/42442 https://thecoatlessprofessor.com/programming/cpp/r-compiler-tools-for-rcpp-on-macos/
After installing the package, please run the following example code:
library(fuzzymineR)
# Load the sample dataset
data("artificial_loan_process")
# Create an eventlog object
log <- create_eventlog(artificial_loan_process,
case_id = "case",
activity_id = "event",
timestamp = "completeTime")
# Mine the fuzzy model
metrics <- mine_fuzzy_model(log, verbose = TRUE)
# Visualize the fuzzy model for a given set of
# parameters
viz_fuzzy_model(metrics = metrics,
node_sig_threshold = 0,
edge_sig_threshold = 0.3,
edge_sig_to_corr_ratio = 0.75)Pleae read the documentation or the research paper linked below for understanding the different metrics in the viz_fuzzy_model function.
Fuzzy Mining - Adaptive Process Simplification Based on Multi-Perspective Metrics
- Tirth Shah (Playpower Labs)
- Nirmal Patel (Playpower Labs)