This repository serves as a reproducibility package, enabling the execution of experiments outlined in the research paper: "Discovery and Simulation of Data-Aware Business Processes"
It includes the necessary source code, datasets, models, and execution guidelines, joining the four necessary systems/libraries into one repository for convenience. The extended systems referenced in the papers are Simod, pix-framework and Prosimos, while the log-distance-measures library provides the metrics used in the evaluation. For the most recent updates and full functionalities, please follow the provided links to access the source codes.
- Python 3.9
- Use Poetry
-
Install the Poetry by following the instructions in the links in the Requirements subsection above.
-
Clone this repository. Move to the project folder, install the dependencies by running the following command:
poetry install -
To run the experiments consider using this commands:
poetry run data-attributesto run data attributes evaluation (synthetic logs)poerty run branching-conditionsto run branching condition evaluation (synthetic logs)poetry run real-lifeto run evaluation for real life logs (BPIC2019, Sepsis and Trafic cases)
Results reported in the paper are available here experiment_results.xlsx
The script first generates 20 logs for each attribute type (global and local) and for multiple or single change within the simulation + log with 10 case attributes
All logs will be saved under out/data_attributes folder. Then for each log it runs data attribute discovery function and appends metric results to the corresponding "metric result" file.
metrics_continuous.csv will contain all results for global/event attributes with numerical values, while metrics_discrete.csv will contain results for categorical results. metrics_case.csv contain both numerical and categorical results, but only for case attributes
Description of each generator used for branching conditions can be found here generate_data_attributes.py
The script first generates 80 configuration files (20 for XOR with short-term dependencies and 20 with long-term dependencies), then does the same for OR gateway. Next we run simulation to create original log (giving 80 logs). Next we split original log into train and test logs, run the discovery on train log. Based on received results we simulate 1 log with traditional approach (without conditions) and second with Data-Aware Simulation model and compare received logs to test log. For comparison we use N-Gram distance (n=3) comparing the received control-flow between pairs original-traditional and original-conditional models.
Description of each generator used for branching conditions can be found here for short-term dependencies generate_branching_conditions_short.py and for long-term dependencies generate_branching_conditions_long.py
Metrics for each run will be save in a single file (long and short term dependencies separately) at out/branching_conditions/long_term_dependencies/simulation_metrics.csv
Here we use already discovered bpmn model and configuration file with Simod (saved in assets). We split the log into train and test, discover attributes and branching conditions using train log, updating the copy of the discovered traditional log and insert there Data-Aware Model components. Then we simulate 2 logs as we did with Branching Condition tests.
Data attribute results will be available for each model separately under real-life log name BPIC2019, Sepsis, Trafic in out/real_life folder and split in the same way as for Data Attribute tests.