Following is the source code for the article "Prediction-based Resource Allocation using LSTM and maximum flow and minimum cost algorithm" by Gyunam Park and Minseok Song presented at the 1st International Conference on Process Mining and submitted to the information systems (special issue).

The code provided in this repository can be readily used to optimize resource scheduling in non-clairvoyant online job shop environment

Requirements:

• This code is written in Python3.6. In addition, you need to install a few more packages.

  • networkx
  • numpy
  • pandas
  • keras
  • tensorflow
  • PyProM

• To install,

  $ cd prediction_based_resource_allocation
  $ pip install -r requirements.txt
  $ cd ..
  $ git clone https://github.com/gyunamister/PyProM.git
  

Implementation:

• Evaluation
  • On artificial log,
    • experiments for RQ1 and RQ2: Type $ sh exp_1.sh to optimize resource allocation with the proposed method and $ sh exp_1-baseline.sh to optimize resource allocation with the baseline approach.
    • experiments for RQ3: Type sh exp_2.sh to optimize resource allocation with the proposed method by varing the prediction accuracy.
  • On real-life log,
    • experiments for RQ1 and RQ2: Type $ sh exp_3.sh to optimize resource allocation with the proposed method and $ sh exp_3-baseline.sh to optimize resource allocation with the baseline approach.
• Brief Explanation
  • The logs are listed in "./sample_data". The artificial log is generated by simulating a business process in an emergency department. Detailed implementation for the simulation is in data/log_generator.py
  • Phase 1 of our method (prediction model construction) is implemented in "prediction/model.py". Configuration is set on "prediction/config.py" and the training is done by python train.py. The resulting prediction model is saved in a directory "./prediction/checkpoints".
  • Phase 2 of our method (prediction model construction) is implemented in optimizer/suggested.py, while the baseline method is implemented in optimizer/baseline.py