Synthetic Event Log Generator

This tool generates synthetic event logs for process mining and process intelligence use cases. It is fully YAML-driven, supports strict business-rule-compliant process variants and deviations, and allows attribute-based assignment for both variants and deviations.

Features

• YAML configuration for all process logic, variants, deviations, durations, and attributes
• Attribute-based assignment: Link variants and deviations to specific case attributes (e.g., CustomerType, Region)
• Multi-step deviations: Deviations can be a sequence of activities
• Strict conformance: No legacy or unreachable code; all logic is YAML-driven
• Precise deviation and variant frequencies (per-case probability/frequency)
• Support for case and event attributes
• Activity and deviation durations (affecting timestamps)
• CSV output for easy analysis

Requirements

• Python 3.7+
• pm4py
• pandas
• pyyaml

Install dependencies with:

pip install pm4py pandas pyyaml

Usage

1. Prepare your YAML configuration

Edit variables.txt to define your process, deviations, attributes, and variants. Example:

context: Order-to-Cash process for B2B sales

activities:
  - Receive Order
  - Check Credit
  - Approve Order
  - Order Change Requested
  - Order Change Approved
  - Re-pick Items
  - Re-pack Items
  - Pick Items
  - Pack Items
  - Partial Shipment
  - Notify Customer
  - Backorder Created
  - Ship Backorder
  - Ship Order
  - Send Invoice
  - Payment Reminder Sent
  - Late Payment
  - Apply Late Fee
  - Receive Payment
  - Complaint Received
  - Return Processed
  - Refund Issued
  - VIP Approval
  - Gift Packaging
  - Credit Check Escalated
  - Manual Review
  - Order Cancellation
  - Express Handling
  - Priority Picking
  - Priority Packing

activity_durations:
  Default: 1440
  Receive Order: 60
  Check Credit: 240
  # ...

case_attributes:
  CustomerType: [New, Returning, VIP]
  Region: [North, South, East, West]

event_attributes:
  Resource: [Alice, Bob, Carol, Dave]
  Channel: [Email, Phone, Web]

variants:
  - name: Standard
    sequence: [Receive Order, Check Credit, Approve Order, Pick Items, Pack Items, Ship Order, Send Invoice, Receive Payment]
    frequency: 0.6
  - name: VIP
    sequence: [Receive Order, Check Credit, Approve Order, VIP Approval, Gift Packaging, Pick Items, Pack Items, Ship Order, Send Invoice, Receive Payment]
    frequency: 0.1
    attributes:
      CustomerType: VIP
  # ...

deviations:
  - name: Late Payment
    probability: 0.15
    after: Send Invoice
    steps: [Payment Reminder Sent, Late Payment, Apply Late Fee]
    description: Customer pays invoice late
    attributes:
      CustomerType: New
      Region: East
  # ...

2. Run the generator

python generate_synthetic_event_log.py --variables-file variables.txt --output-file synthetic_event_log.csv --n-cases 1000 --random-seed 42 --max-activities 20

Key CLI options

• --variables-file: Path to YAML configuration (default: variables.txt)
• --output-file: Output CSV file (default: synthetic_event_log.csv)
• --n-cases: Number of cases to generate (default: 100)
• --random-seed: Set for reproducible results
• --max-activities: Maximum number of activities per case (default: 8)

3. Validate and analyze the output

Use the unified QA script to check conformance, ratios, and summary:

python qa_event_log.py --all

Or run individual checks:

python qa_event_log.py --strict
python qa_event_log.py --summary
python qa_event_log.py --ratios

4. Output

The generated CSV will contain one row per event, with columns for case name, activity, timestamp, and all configured attributes.

Advanced Configuration

• Attribute-based assignment: Use the attributes field in variants or deviations to restrict them to specific case attribute values.
• Multi-step deviations: Use the steps field to insert a sequence of activities for a deviation.
• Strict placement: Deviations are only inserted if the placement rule matches the variant path; otherwise, a warning is logged and the deviation is skipped.
• Timestamps and durations: Each activity (including deviation steps) gets its own duration and timestamp, reflecting the true process flow and impact of deviations.
• Variants: Frequencies should sum to 1.0. Cases are distributed accordingly.
• Attributes: Add as many case/event attributes as needed; values are randomly assigned unless restricted by a variant or deviation.

Example Output

case:concept:name	Activity	Timestamp	CustomerType	Region	Resource	Channel
Case_1	Receive Order	2025-07-27 10:00:00	VIP	North	Alice	Email
Case_1	Check Credit	2025-07-27 10:10:00	VIP	North	Bob	Phone
Case_1	Approve Order	2025-07-27 10:20:00	VIP	North	Carol	Web
Case_1	Order Change Requested	2025-07-27 10:25:00	VIP	North	Dave	Email
...	...	...	...	...	...	...

License

MIT License

Author

Nick Blackbourn