Scalable process discovery and conformance checking.

Area of Science:

• Computer Science
• Data Mining
• Process Mining

Background:

• Organizations collect vast amounts of event data, necessitating efficient process discovery techniques.
• Existing process discovery methods often lack scalability or strong quality guarantees for large datasets.
• There is a need for techniques that can handle billions of events and thousands of activities while ensuring model soundness.

Purpose of the Study:

• To introduce a novel framework for process discovery that ensures scalability and quality.
• To develop algorithms that can discover sound process models from large event logs.
• To establish a robust framework for measuring the quality of discovered process models.

Main Methods:

• A new framework for process discovery is proposed, designed for single-pass log traversal.
• Three algorithms are introduced that leverage this framework for process discovery.
• A divide-and-conquer strategy is employed for model-log and model-model comparison to assess quality metrics like recall, fitness, and precision.

Main Results:

• The proposed framework and algorithms demonstrate scalability, handling logs with 100,000,000 traces and processes with 10,000 activities on standard hardware.
• The discovered models maintain high quality, with minimal sacrifice compared to other algorithms.
• The techniques ensure the rediscovery of underlying processes with sufficient data.

Conclusions:

• The developed framework offers a scalable and high-quality solution for process discovery from large event logs.
• This approach addresses the limitations of existing methods in handling big data in process mining.
• The study contributes efficient algorithms and quality measurement techniques for practical process mining applications.