Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Novel Approach to Process Discovery with Enhanced Loop Handling

Ali Nour Eldin, Benjamin Dalmas, Walid Gaaloul

TL;DR

Bonita Miner tackles the challenge of process discovery in logs with complex loop structures by integrating Directly-Follows Graph filtering and a Depth-First Algorithm driven gateway construction, with loops built separately to keep models simple yet behaviorally accurate. The two-phase approach first purges concurrency and loops from the graph, then constructs a BPMN diagram in a single pass, producing smaller, more interpretable models that balance fitness, precision, and generalization. Empirical results on BPIC real logs and synthetic loop scenarios show Bonita Miner achieving high fitness and F-scores while reducing model complexity compared to state-of-the-art baselines, demonstrating robust loop handling and generalization. The work offers practical BPMN models with improved readability and reliability, and points to future work in runtime optimization and deadlock detection to further increase industrial applicability.

Abstract

Automated process discovery from event logs is a key component of process mining, allowing companies to acquire meaningful insights into their business processes. Despite significant research, present methods struggle to balance important quality dimensions: fitness, precision, generalization, and complexity, but is limited when dealing with complex loop structures. This paper introduces Bonita Miner, a novel approach to process model discovery that generates behaviorally accurate Business Process Model and Notation (BPMN) diagrams. Bonita Miner incorporates an advanced filtering mechanism for Directly Follows Graphs (DFGs) alongside innovative algorithms designed to capture concurrency, splits, and loops, effectively addressing limitations of balancing as much as possible these four metrics, either there exists a loop, which challenge in existing works. Our approach produces models that are simpler and more reflective of the behavior of real-world processes, including complex loop dynamics. Empirical evaluations using real-world event logs demonstrate that Bonita Miner outperforms existing methods in fitness, precision, and generalization, while maintaining low model complexity.

A Novel Approach to Process Discovery with Enhanced Loop Handling

TL;DR

Bonita Miner tackles the challenge of process discovery in logs with complex loop structures by integrating Directly-Follows Graph filtering and a Depth-First Algorithm driven gateway construction, with loops built separately to keep models simple yet behaviorally accurate. The two-phase approach first purges concurrency and loops from the graph, then constructs a BPMN diagram in a single pass, producing smaller, more interpretable models that balance fitness, precision, and generalization. Empirical results on BPIC real logs and synthetic loop scenarios show Bonita Miner achieving high fitness and F-scores while reducing model complexity compared to state-of-the-art baselines, demonstrating robust loop handling and generalization. The work offers practical BPMN models with improved readability and reliability, and points to future work in runtime optimization and deadlock detection to further increase industrial applicability.

Abstract

Automated process discovery from event logs is a key component of process mining, allowing companies to acquire meaningful insights into their business processes. Despite significant research, present methods struggle to balance important quality dimensions: fitness, precision, generalization, and complexity, but is limited when dealing with complex loop structures. This paper introduces Bonita Miner, a novel approach to process model discovery that generates behaviorally accurate Business Process Model and Notation (BPMN) diagrams. Bonita Miner incorporates an advanced filtering mechanism for Directly Follows Graphs (DFGs) alongside innovative algorithms designed to capture concurrency, splits, and loops, effectively addressing limitations of balancing as much as possible these four metrics, either there exists a loop, which challenge in existing works. Our approach produces models that are simpler and more reflective of the behavior of real-world processes, including complex loop dynamics. Empirical evaluations using real-world event logs demonstrate that Bonita Miner outperforms existing methods in fitness, precision, and generalization, while maintaining low model complexity.

Paper Structure

This paper contains 20 sections, 1 equation, 11 figures, 10 algorithms.

Table of Contents

  1. Introduction
  2. Related work
  3. Approach Overview
  4. Preliminary
  5. Bonita Miner
  6. Running Example
  7. Concurrency discovery
  8. Loop filtering
  9. Synthesis of Control Flow Elements
  10. Exclusive & Concurrency Relations Analysis
  11. Loop Structures Analysis
  12. Split & Join Discovery
  13. Split Operation
  14. Join Operation
  15. Loop Construction
  16. ...and 5 more sections

Figures (11)

  • Figure 1: A simple example of the limitation of existing works.
  • Figure 2: Bonita Miner Overview
  • Figure 3: The running example is used to explain the Bonita Miner workflow
  • Figure 4: Part of the DFG of the event log of the running example
  • Figure 5: DFG, after filtering the parallelism, of the running example with cycles
  • ...and 6 more figures

Theorems & Definitions (9)

  • definition thmcounterdefinition: Event Log
  • definition thmcounterdefinition: Directly-Follows Graph
  • definition thmcounterdefinition: Path
  • definition thmcounterdefinition: Cycle
  • definition thmcounterdefinition: BPMN Process Model
  • definition thmcounterdefinition: Concurrent Relation
  • definition thmcounterdefinition: Branching Parallel Relations
  • definition thmcounterdefinition: Looping edge
  • definition thmcounterdefinition: Branching Exclusive Relations