This project is the implementation of the Decay Replay Mining - Next Transition Prediction (DREAM-NAP) approach described in the paper Decay Replay Mining to Predict Next Process Events by Julian Theis and Houshang Darabi. Both authors are part of the Process Mining and Intelligent System Analytics Team (PROMINENT) at the University of Illinois at Chicago, USA.
The DREAM-NAP approach consists of two stages: Decay Function Enhancement of a process model and Deep Learning using timed state samples.
The folder DREAM contains the Java source code from process discovery to extration of timed state samples whereas NAP contains the Python files to train and evaluate the neural network.
Bold values designate that the proposed model outperforms state-of-the-art results.
∗ denotes datasets that do not contain resources, therefore DREAM-NAPr is not applicable.
∗∗ denotes that the source code of Breuker et al. [12] was not able to produce results on this dataset.
The PyDREAM project is a Python-based implementation of DREAM and DREAM-NAP using the open source process mining platform PM4Py. The documentation of the project and examples can be found in the corresponding Github repository: https://github.com/Julian-Theis/PyDREAM.
We have implemented the Decay Replay Mining (DREAM) preprocessing approach as a ProM plugin. The plugin considers a Petri net process model as PNML and an CSV formatted event log as input and produces timed state samples that can be used for further machine learning and data science tasks. The plugin enhances and parametrizes each place of the process model with a time decay function. Afterwards, the event log is replayed on the enhanced model and timed state samples are extracted at every discrete timestep observed in the log.
The plugin and its documentation is available here: https://prominentlab.github.io/ProM-DREAM/.
@article{theis2019decay,
title={Decay Replay Mining to Predict Next Process Events},
author={Theis, Julian and Darabi, Houshang},
journal={IEEE Access},
volume={7},
pages={119787--119803},
year={2019},
publisher={IEEE}
}
We thank Raffaele Conforti for making his extensive research code available. This project requires the installation of his Research Code.