Libra is a framework that leverages recently developed Privacy-Enhancing Technologies (PETs) to help organizations achieve a better privacy-utility trade-off while publishing anonymized process models. Libra uses the idea of privacy amplification to reduce the injected noise while providing differential privacy guarantees. The proposed method could be adapted with any event log anonymization technique to enhance the utility.
The main dependencies are: pm4py, diffprivlib, multiprocessing and statistics You can install all the requirements with:
pip install -r requirements.txt
Libra is available as a python package. To anonymize an event log, place the XES file in the directory data. Then you can run the command
python Libra.py "BPIC12_t" 2 0.05 10 20 0.001
The above parameters are : event log, b, gamma, alpha, epsilon_in_minutes, delta.
Libra assumes that the event log has only the three columns: case:concept:name, concept:name, and time:timestamp in your XES file.
For more information about the description of these parameters, please check out the paper.
@inproceedings{DBLP:conf/icpm/ElkoumyD22,
author = {Gamal Elkoumy and
Marlon Dumas},
editor = {Andrea Burattin and
Artem Polyvyanyy and
Barbara Weber},
title = {Libra: High-Utility Anonymization of Event Logs for Process Mining
via Subsampling},
booktitle = {4th International Conference on Process Mining, {ICPM} 2022, Bolzano,
Italy, October 23-28, 2022},
pages = {144--151},
publisher = {{IEEE}},
year = {2022},
url = {https://doi.org/10.1109/ICPM57379.2022.9980619},
doi = {10.1109/ICPM57379.2022.9980619},
timestamp = {Tue, 03 Jan 2023 07:35:57 +0100},
biburl = {https://dblp.org/rec/conf/icpm/ElkoumyD22.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}