Amun

Amun is a framework that leverage recently developed Privacy-Enhancing Technologies (PETs) to help organizations publish anonymized process models. The main contributions of the project are listed below.

• Amun proposes a framework to anonymize event logs to the extent that no individual can be singled out using the anonymized log. It anonymizes event logs in order to guarantee that, upon the disclosure of the anonymized log, the probability that an attacker may single out any individual represented in the original log, does not increase by more than a threshold. Amun proposes a differentially private disclosure mechanism, which oversamples the cases in the log and adds noise to the timestamps to the extent required to achieve the above privacy guarantee. An emperical evaluation of Amun using 14 real-world event logs can be found in our recent paper. Also, Amun supports other anonymization approaches such as sampling and filtering, which are presented in the journal extension paper.
• Amun uses a mathematically proven privacy model to balance the risk correlated with publishing process models and the utility after anonymization. It uses an ε-differential-privacy mechanism to anonymize Directly-Follows Graphs (DFGs). It provides a mathematical approach to calculate the value of ε that represents the amount of noise injected a process mining model that optimizes the risk and utility measures. Amun keeps all the traces and all the activities of a DFG. An emperical evaluation of Amun using 13 real-world event logs could be found in our recent paper.

Availability

Amun is available as a python package and a docker image. To anonymize an event log, place the XES file in the directory input_logs. Then you can run the command

python Amun.py Sepsis sampling 0.2

Amun assumes that the event log has only the three columns: case:concept:name, concept:name, and time:timestamp in your XES file.

Docker Image

The docker image and the installation steps are presented at the branch amun-flask-app.

Prerequisite

The main dependencies are: pm4py, diffprivlib, multiprocessing and statistics You can install all the requirements with:

pip install -r requirements.txt

The code was tested with python 3.8.5.

Reproduce Emperical Evaluation

An example of the usage of Amun to anonymize DFGs can be found in the file run_example.py. To perform the execution time experiment, you can use the file run_example_execution_time.py. An implementation of Amun with a single thread could be found in the branch amun-model.

To reproduce the emperical evaluation of Amun to anonymize event logs you can use the file run_event_log_anonymizer.py.

Cite the Project

Our Information Systems paper:

@article{ELKOUMY2022102161,
title = {Differentially private release of event logs for process mining},
journal = {Information Systems},
pages = {102161},
year = {2022},
issn = {0306-4379},
doi = {https://doi.org/10.1016/j.is.2022.102161},
url = {https://www.sciencedirect.com/science/article/pii/S0306437922001399},
author = {Gamal Elkoumy and Alisa Pankova and Marlon Dumas},
}

Our ICPM22 Demo paper:

@inproceedings{DBLP:conf/icpm/ElkoumyPD22,
  author    = {Gamal Elkoumy and
               Alisa Pankova and
               Marlon Dumas},
  title     = {Amun: {A} tool for Differentially Private Release of Event Logs for
               Process Mining (Extended Abstract)},
  booktitle = {{ICPM} Doctoral Consortium / Demo},
  series    = {{CEUR} Workshop Proceedings},
  volume    = {3299},
  pages     = {56--60},
  publisher = {CEUR-WS.org},
  year      = {2022}
}

Our ICPM21 paper:

@inproceedings{DBLP:conf/icpm/ElkoumyPD21,
  author    = {Gamal Elkoumy and
               Alisa Pankova and
               Marlon Dumas},
  editor    = {Claudio Di Ciccio and
               Chiara Di Francescomarino and
               Pnina Soffer},
  title     = {Mine Me but Don't Single Me Out: Differentially Private Event Logs
               for Process Mining},
  booktitle = {3rd International Conference on Process Mining, {ICPM} 2021, Eindhoven,
               Netherlands, October 31 - Nov. 4, 2021},
  pages     = {80--87},
  publisher = {{IEEE}},
  year      = {2021},
  url       = {https://doi.org/10.1109/ICPM53251.2021.9576852},
  doi       = {10.1109/ICPM53251.2021.9576852},
  timestamp = {Fri, 29 Oct 2021 16:42:41 +0200},
  biburl    = {https://dblp.org/rec/conf/icpm/ElkoumyPD21.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}
  

Undersubmission preprint:

@article{DBLP:journals/corr/abs-2012-01119,
  author    = {Gamal Elkoumy and
               Alisa Pankova and
               Marlon Dumas},
  title     = {Privacy-Preserving Directly-Follows Graphs: Balancing Risk and Utility
               in Process Mining},
  journal   = {CoRR},
  volume    = {abs/2012.01119},
  year      = {2020},
  url       = {https://arxiv.org/abs/2012.01119},
  eprinttype = {arXiv},
  eprint    = {2012.01119},
  timestamp = {Fri, 04 Dec 2020 12:07:23 +0100},
  biburl    = {https://dblp.org/rec/journals/corr/abs-2012-01119.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}