Process Mining with PM4Py

A Hands-On Tutorial Using Real Insurance Claims Event Log Data

This project demonstrates how to perform process discovery, bottleneck detection, and visual analysis using the PM4Py process mining library in Python.
The notebook processes an insurance claims event log, converts it into an event log structure, and generates multiple Directly-Follows Graphs (DFGs).

---

Project Overview

This notebook covers:

• Loading and preparing a real event log dataset (insurance claims)
• Converting a CSV file into a PM4Py-compatible event log
• Discovering the process model using:
  • Standard DFG
  • Frequency-based DFG
  • Performance-based DFG
• Visualizing each discovered model
• Understanding process flow and identifying inefficiencies

---

Dataset Description

The dataset contains dummy HR onboarding events with fields including:

• Employee_ID — unique claim identifier
• Activity — process step
• timestamp — event time

These were mapped to PM4Py event-log fields: Employee_ID is mapped to case:concept:name Activity is mapped to concept:name timestamp is mapped to time:timestamp

---

Technologies Used

• Python 3
• PM4Py – Process Mining for Python
• Pandas – Data handling
• Google Colab – Runtime environment

---

Installation

Inside the notebook, PM4Py is installed with:

!pip install pm4py

Import Libraries & Load Dataset

Load your CSV event log and inspect the initial data.

import pm4py

import pandas as pd

df = pd.read_csv('HR_Onboarding.csv')

Convert DataFrame → PM4Py Event Log

Convert your cleaned DataFrame into a PM4Py event log.

df = df.rename(columns={
    "Employee_ID": "case:concept:name",
    "Activity": "concept:name",
    "Timestamp": "time:timestamp"
})
df["time:timestamp"] = pd.to_datetime(df["time:timestamp"])

Discover Standard Directly-Follows Graph (DFG)

Generate and visualize the initial DFG.

import pm4py
from pm4py.algo.discovery.dfg import algorithm as dfg_discovery
from pm4py.visualization.dfg import visualizer as dfg_visualization
log = pm4py.convert_to_event_log(df)
dfg = dfg_discovery.apply(log)

Standard DFG Visualization

gviz = dfg_visualization.apply(dfg, log=log)
dfg_visualization.view(gviz)

It will display image

---

Frequency-Based DFG

from pm4py.algo.discovery.dfg import algorithm as dfg_discovery
from pm4py.visualization.dfg import visualizer as dfg_visualization

dfg_freq = dfg_discovery.apply(log, variant=dfg_discovery.Variants.FREQUENCY)
gviz_freq = dfg_visualization.apply(dfg_freq, log=log, variant=dfg_visualization.Variants.FREQUENCY)
dfg_visualization.view(gviz_freq)

Performance-Based DFG

dfg_perf = dfg_discovery.apply(log, variant=dfg_discovery.Variants.PERFORMANCE)
gviz_perf = dfg_visualization.apply(dfg_perf, log=log, variant=dfg_visualization.Variants.PERFORMANCE)
dfg_visualization.view(gviz_perf)

Repository Structure

| mining.ipynb is Main Google Colab notebook | | HR_Onboarding.csv is Input event log dataset | | dfg_frequency.png is Frequency DFG image | | dfg_performance.png is Performance DFG image | | README.md is Project documentation |

Purpose of the Notebook

This notebook helps you:

• Understand event logs and their structure
• Transform raw process data into PM4Py format
• Discover process models using DFGs
• Analyze bottlenecks using performance metrics
• Build a reproducible process mining workflow

---

How to Use This Notebook

1. Open the notebook in Google Colab
2. Upload the event log CSV
3. Run all cells sequentially
4. Visualizations (DFGs) will appear automatically

The HR Onboarding dataset contains:

• Employee_ID
• Activities
• Timestamps

These are converted into PM4Py’s event log format during preprocessing.

---

References

• PM4Py Documentation

  Author Satish Gupta

This project was created as part of a hands-on learning exercise in Process Mining with Python & PM4Py. Check project

!pip install pm4py