/BusinessProcessOutcomePrediction

Labpackage for paper results

Primary LanguagePython

Repository for the results presented in "Considering Control Flow Constructs for Predicting Business Process Outcomes with Deep Learning"

Content

The repo contains three folders:

  • code: contains the python machine learning code to train and evaluate the models
  • data: contains both raw and transformed data of the used dataset
  • transform: little .NET 4.5 application that was used to transform the data in the ./data/ folder

Install

To run the python code you need the following:

  • Python 2
  • Keras
  • Backend of your choice (we used CNTK)
  • GPU: cuda installation and cuda compatible backend

and the following python packages:

  • unicodecsv
  • distance
  • jellyfish

If you want to use docker, there are ready-to-use images in chemsorly/keras-cntk

Run

  • Clone repository
  • Navigate to ./code/
  • run "python s2s.py 1 100 0.1 20 1"

Parameters are as follows:

  • Running number (unused int)
  • Neurons per layer (int)
  • Dropout (double 0-1)
  • Patience (int)
  • Optimization algorithm (int 1-7)

Credits

The Cargo 2000 Freight Tracking and Tracing Data Set is available at UCL (Citation)
Credits go to verenich whose work was used as base for this project.

Reference

(tba)