README.md

CONTENTS

• Introduction
• Process Mining Pipeline: Workflow
• Stack & Installation/Configuration Requirements
• Maintainers
• Backlog (Shared Doc Link)
• Troubleshooting

INTRODUCTION

• This is a Proof of Concept (PoC) for distributed Process Mining (PM) data pipelines. It relies on the theoretical concepts of:

1. graphs:
   • Directly-Follows Graph (DFG)
   • Directed Acyclic Graph (DAG)
2. database:
   • Neo4J graph db (via database .jar connector)
3. data analysis:
   • distributed analytical engine, i.e., Apache Spark
     • interface: PySpark
   • PM4Py Process Mining algorithms (Python-based package)

PROCESS MINING PIPELINE: WORKFLOW

• log files, i.e., .csv files
  • Apache Spark via PySpark interface
    • data partitioning (by case ID) and time windowing (by timestamps)
      • Directly-Follows Graph schema, i.e., predecessor(s) and successor(s) nodes
        • WRITE queries to a Neo4J graph database
          • format conversion for algorithmic analysis, i.e., Parent/Child nodes and their frequency
            • derive Petri net objects and put them within dataframes
              • PM4Py analysis by applying - in parallel to DFGs - Process Mining algorithms, i.e.,
                1. α-miner
                2. heuristic miner
                3. inductive miner
                   • plotting of evaluation metrics via comparative charts from the dataframes - derived by converting Resilient Distributed Database (RDD)s to Pandas

STACK & INSTALLATION/CONFIGURATION REQUIREMENTS

System config:

• pySpark vs. 2.4.6
• Java JVM vs. 8
• Scala vs. 2.13

YAML file for setting Conda and Jupyter notebook environment:

name: pyspark-vs2
channels:
  - defaults
dependencies:
  - pip=20.2.4
  - python=3.7.9
  - pip:
    - matplotlib==3.3.4
    - numpy==1.20.0
    - pandas==1.2.1
    - pm4py==2.2.5
    - py4j==0.10.9.1
    - pydotplus==2.0.2
    - pyspark==2.4.7
    - python-graphviz==0.8.4

NOTE: if some packages are missing in the required version, try installing from pip or from Conda-forge: $ conda install -c conda-forge [pkg_name]

CONTRIBUTORS/MAINTAINERS

• Prof. Dr. Ahmed Awad [email protected]
• Belal Mohammad: [email protected]
• Fabiano Spiga: [email protected]

BACKLOG (SHARED DOC LINK)

Backlog (Shared Doc)