Conditioning & Stability Analysis of Process Discovery Algorithms

This repository contains the full experimental codebase used to analyze the conditioning and stability of process discovery algorithms. The workflow is organized into four modular sub-projects, each responsible for a key stage of the experimental pipeline.

Project Structure

1. LogGeneration

Generates synthetic event logs for experimentation.

• Uses SNIP to inject controlled noise.
• Supports multiple noise types and parameter configurations.
• Produces input logs for downstream discovery tasks.

To run, update src/main/resources/systems/ folder with the systems folder in https://doi.org/10.26188/30739082, and run src/main/java/org/anandi/Main.java. The clean samples from the base logs and noisy versions of those clean samples will be generated in src/main/resources/logs/.

2. ModelDiscovery

Discovers process models for each generated log.

• Implements Alpha Miner, Heuristics Miner, and Inductive Miner.
• Provides consistent interfaces for batch model discovery.
• Outputs models in standardized formats for evaluation.

To run, run model_discovery.py (if needed, update the base file path to where noisy logs are stored). The code assumesa path <base_path>/logs/. Models will be generated at <base_path>/models/. The clean models are discovered first, then noisy models are discovered in a random order (according to log_data.csv).

3. Evaluation

Computes the core measures needed for conditioning and stability analysis.

• Quantifies differences between: logs; discovered models; behavioral perspectives; structural perspectives.
• Produces structured, machine-readable evaluation results.

For rediscoverability analysis, run evaluation_f1.py. If needed, update the paths to systems and models. For behavior replay analysis, run evaluation_f2.py. If needed, update the paths to logs and models.

4. Analysis

Analyzes conditioning and stability results using the computed metrics.

• Performs statistical comparisons and robustness assessments.
• Generates all visualizations used in the study (plots, summaries, and tables).
• Includes scripts for reproducing final figures.

For conditioning (model complexity changes with log changes), run conditioning.py, to generate the analysis report and plots. For stability analysis: (1) for rediscoverability analysis, run stability_f1.py, (2) for behavior replay analysis, run stability_f2.py, to generate the analysis report and plots.