Meta-learning for Anomaly Detection in Process Mining

This file lists the steps to reproduce the experiments, analysis and figures generated for the paper Process Mining Encoding via Meta-learning for an Enhanced Anomaly Detection.

Contents

This repository already comes with the datasets employed in the experiments along with the code to reproduce them. We also provide the experimental results (see .csv files) and figures used in the papers (see plots folder).

Installation steps

First, you need to install conda to manage the environment. See installation instructions here.

The next step is to create the environment. For that, run:

conda create --name mtl_anomaly python=3.7.0

Then, activate the environment:

conda activate mtl_anomaly

Finally, install the dependencies:

python -m pip install -r requirements.txt

conda install -c conda-forge openjdk

Reproducing experiments

The first step is to extract the meta-features from the event logs:

python compute_encoding/extract_meta_features.py

Then, we compute the encodings of each method:

python compute_encoding/doc2vec.py
python compute_encoding/node2vec_.py
python compute_encoding/alignment.py

The encodings are then submitted to a classification pipeline and metrics regarding their performance are recorded:

python compute_encoding/classification.py

The last step creates the analysis and plots the results in a dedicated folder:

python compute_encoding/analysis.py