Skip to content

Latest commit

 

History

History
78 lines (64 loc) · 3.4 KB

README.md

File metadata and controls

78 lines (64 loc) · 3.4 KB
Raw

A deep-learning-based method for detecting workarounds in event log data

How to use the method?

  1. Create a virual environment (you find further details about that here)
  2. Install the required python packages through the command "pip install -r requirements.txt" (requirements.txt)
  3. Set relevant parameters in the file "config.py" (wdwdl/config.py)
  4. Execute the file "main.py" (wdwdl/main.py)

Steps of the method

  1. Split event log in data set and prediction set (90%/10%)
  2. Filter out noise from data set by using an Autoencoder (AE) model
  3. Integration of workarounds (rules enriched by random elements, 30% of the instances include a workaround)
  4. Learn a convolutional neural network (CNN) model for detecting workarounds
    • Based on 80% of the data set
    • Test model on 20% of the data set
    • Hyper-parameter optimization (HPO) with a 10%-validation-set of the training set
  5. Apply the CNN model on the prediction set (10%) that includes noise

Setting

  • Metrics (note: we use the macro version of Precision, Recall and F1-Score to better consider workaround classes with fewer instances):
    • Accuracy,
    • Precision (macro),
    • Recall (macro),
    • F1-score (macro)
  • Encoding: binary encoded categorical attributes (note: we binary encode the categorical attributes because this encoding is less sparse than one-hot encoding)
  • Validation method: split validation
  • Shuffling:
    • For general data set split (data set and pred set): yes
    • For data set split (train test): yes
    • For hyper-parameter optimisation: yes
    • For validation set per epoch (to perform early stopping): yes
  • Hyper-parameter optimisation (HPO):
    • For the AE model: Tree-Structured Parzen Estimators approach (TPE)
    • For the CNN model: Tree-Structured Parzen Estimators approach (TPE)
    • Number of HPO runs: 10
    • Optimisation criteria: F1-score
  • Data augmentation (i.e. replace noisy process instances by normal process instances): yes
  • Threshold determination: depending on the event log the x'th percentile where x = {50, 60, 70, 80, 90, 100}

Paper

This code repository belongs to a paper published in the Journal of Business Analytics. If you use this code, please cite the paper as follows.

@article{weinzierl2021wdwdl,
    title={Detecting Temporal Workarounds in Business Processes - A Deep-Learning-based Method for Analysing Event Log Data},
    author={Sven Weinzierl and Verena Wolf and Tobias Pauli and Daniel Beverungen and Martin Matzner},
    journal={Journal of Business Analytics},
    volume = {5},
    number = {1},
    pages = {76-100},    
    year={2022},
    publisher={Taylor & Francis},
    doi = {10.1080/2573234X.2021.1978337}
}

You can access this paper here.

A previous version of the JBA paper was published in the proceedings of the 28th European Conference on Information Systems (ECIS2020).

@proceedings{weinzierl2020wdwdl,
    title={Detecting Workarounds in Business Processes - a Deep Learning method for Analyzing Event Logs},
    author={Sven Weinzierl and Verena Wolf and Tobias Pauli and Daniel Beverungen and Martin Matzner},
    booktitle={Proceedings of the 28th European Conference on Information Systems (ECIS2020)},
    year={2020},
    pages={1-16}
    organization={AISeL}
}

You can access this paper here.