Quantifying Explainability in Outcome-Oriented Predictive Process Monitoring

Complementary code to reproduce the work of Quantifying Explainability in Outcome-Oriented Predictive Process Monitoring

An overview of the files:

Exploratory Data Analysis and Feature Selection

• EDA_FeatureSelection

Hyperoptimalisation of parameters

• Hyperopt_LogitLeafModel
• Hyperopt_MachineLearningModels

Training of the Machine Learning Models

Logit Leaf Model (LLM), Generalized Logistic Rule Regression (GLRM) and XGBoost)

• Experiment_BPIC2017
• Experiment_TF1
• Experiment_BPIC2015

Training of the Deep Learning Models

Long short-term memory neural networks (LSTM)

• LSTM_BPIC2017
• LSTM_TF1
• LSTM_BPIC2015

Quantitative Metrics

This folder contains the notebook files as listened underneath. Note that for code reproduction of the quantitative metrics, these should be placed outside this folder.

• Experiment_BPIC2017
• Experiment_TF1
• Experiment_BPIC2015
• LSTM_BPIC2017
• LSTM_TF1
• LSTM_BPIC2015

The preprocessing and hyperoptimalisation are derivative work based on the code provided by https://github.com/irhete/predictive-monitoring-benchmark. We would like to thank the authors for the high quality code that allowed to fastly reproduce the provided work. Secondly, we acknowledgde the work provided by https://github.com/renuka98/interpretable_predictive_processmodel architecture to create the long short-term neural networks with attention layers visualisations.

Preprocessing files

• dataset_confs
• DatasetManager
• EncoderFactory

In the Feature Selection file, the original .XES files are used. These can be downloaded from:

• BPIC2017: https://data.4tu.nl/articles/dataset/BPI_Challenge_2017/12696884
• TF1: https://data.4tu.nl/articles/dataset/Road_Traffic_Fine_Management_Process/12683249
• BPIC2015: https://data.4tu.nl/articles/dataset/BPI_Challenge_2015_Municipality_2/12697349/1

Finally, the folders contain additional figures and plots that have not been used in the paper.