Comprehensive Attribute Encoding and Dynamic LSTM HyperModels for Predictive Business Process Monitoring

Predictive Business Process Monitoring (PBPM) research library featuring LSTM-based HyperModels with advanced attribute embeddings. Designed for real-world enterprise process analytics, this toolkit balances flexibility, performance, and reproducibility.

Authors: Fang Wang (Florence Wong), Paolo Ceravolo, Ernesto Damiani
Repository: Code and Demos for the associated research article.

---

📖 Overview

Download Full Paper
This repository provides implementations of attribute encoding techniques and LSTM-based HyperModels for outcome-oriented predictive business process monitoring. The models support multiple scenario setups, including:

• Handling both balanced and imbalanced datasets.
• Utilizing pseudo-embedding matrices for duration and feature correlations.
• Supporting hierarchical inputs for sequence and event attributes.
• Accommodating simultaneous event inputs.

The framework is designed for flexibility and performance across diverse predictive process monitoring tasks.

🧩 Repository Structure

🔧 Embedding and Encoding

File	Description
DurationEmbedding.py	Duration pseudo-embedding matrix and encoding.
FeatureEmbedding.py	Co-relation pseudo-embedding matrix and encoding.
DataEncoder.py	Event-level and sequence-level attribute encoding, including multidimensional encoding.

🤖 HyperModels (LSTM Variants)

Model	File	Description
B-LSTM	BaseLSTM.py	Baseline LSTM for balanced datasets.
B-LSTM (Imbalanced)	BaseLSTMIm.py	Baseline LSTM for imbalanced datasets.
D-LSTM	DurationEmbeddingLSTM.py	LSTM with duration embeddings.
D-LSTM (Imbalanced)	DurationEmbeddingLSTMIm.py	Duration-embedded LSTM for imbalanced data.
DC-LSTM	FeatureDurationEmbeddingLSTM.py	LSTM with feature + duration embeddings.
DC-LSTM (Imbalanced)	FeatureDurationEmbeddingLSTMIm.py	DC-LSTM for imbalanced data.
T-LSTM	TextFeatureDurationEmbeddingLSTM.py	LSTM with text + feature + duration embeddings.
T-LSTM (Imbalanced)	TextFeatureDurationEmbeddingLSTMIm.py	T-LSTM for imbalanced data.

🎛️ Demos (Jupyter Notebooks)

Notebook	Purpose
FeatureEmbedding_call.ipynb	Demo for feature embedding.
DurationBin_call.ipynb	Demo for duration bin encoding.
PatientsBaseLSTM_call.ipynb	Runs B-LSTM (adjustable for balanced/imbalanced datasets).
PatientsTextEmbeddingLSTM_call.ipynb	Runs T-LSTM (adjustable for balanced/imbalanced datasets).
PatientsEmbeddingLSTM_call.ipynb	Runs D-LSTM and DC-LSTM (adjustable for balanced/imbalanced datasets).
BPI12ConcurrLSTM_call.ipynb	Runs M-B-LSTM (balanced dataset demo, adjustable for imbalanced).
BPI12EmbeddingLSTM_call.ipynb	Runs F-D-LSTM (balanced dataset demo, adjustable for imbalanced).

🧠 Coming Soon: HyperGNN for PBPM

We are currently finalizing a self-tuning HyperModel toolkit based on GCN architectures for outcome-oriented predictive business process monitoring (PBPM), extending the current LSTM-based framework. Key features include:

• Multiple GCN variants integrated into a HyperModel structure with industrial-scale, self-tuning hyperparameters.
• Support for diverse graph representations of event sequences with configurable node- and graph-level attributes.
• Temporal dependencies encoded via edge weights to capture sequence dynamics.
• Optimized for prediction accuracy and stability across both balanced and imbalanced datasets.

📌 Code and preprints will be released upon paper acceptance.

---

📜 Citation

If you use this code, please cite the original paper:

@article{Wang20205LSTMHyperPBPM,
         title={Comprehensive Attribute Encoding and Dynamic LSTM HyperModels for Outcome Oriented Predictive Business Process Monitoring}, 
         author={Fang Wang and Paolo Ceravolo and Ernesto Damiani},
         year={2025},
         eprint={2506.03696},
         archivePrefix={arXiv},
         primaryClass={cs.LG},
         url={https://arxiv.org/abs/2506.03696}, 
}

🔗 About the Author

This repository is maintained by Florence Wong, Ph.D. in Business Analytics and Applied Machine Learning. For collaboration, contact via http://www.linkedin.com/in/florence-wong-fw