Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Determining Window Sizes using Species Estimation for Accurate Process Mining over Streams

Christian Imenkamp, Martin Kabierski, Hendrik Reiter, Matthias Weidlich, Wilhelm Hasselbring, Agnes Koschmider

TL;DR

This work addresses the challenge of choosing representative window sizes for streaming process mining under concept drift. It proposes a dynamic window sizing method that uses biodiversity-inspired estimators, notably the Chao1-based species richness estimate, to compute completeness and coverage, coupled with a dynamic elbow-based threshold to close windows. The approach supports multiple species views (activities, trace variants, directly-follows relations) and demonstrates improved accuracy and robustness to drift while maintaining reasonable latency and throughput. Empirical results on synthetic and real-world logs show favorable conformance metrics with linear scaling in the number of observed species, indicating practical applicability in real-time settings.

Abstract

Streaming process mining deals with the real-time analysis of event streams. A common approach for it is to adopt windowing mechanisms that select event data from a stream for subsequent analysis. However, the size of these windows denotes a crucial parameter, as it influences the representativeness of the window content and, by extension, of the analysis results. Given that process dynamics are subject to changes and potential concept drift, a static, fixed window size leads to inaccurate representations that introduce bias in the analysis. In this work, we present a novel approach for streaming process mining that addresses these limitations by adjusting window sizes. Specifically, we dynamically determine suitable window sizes based on estimators for the representativeness of samples as developed for species estimation in biodiversity research. Evaluation results on real-world data sets show improvements over existing approaches that adopt static window sizes in terms of accuracy and robustness to concept drifts.

Determining Window Sizes using Species Estimation for Accurate Process Mining over Streams

TL;DR

This work addresses the challenge of choosing representative window sizes for streaming process mining under concept drift. It proposes a dynamic window sizing method that uses biodiversity-inspired estimators, notably the Chao1-based species richness estimate, to compute completeness and coverage, coupled with a dynamic elbow-based threshold to close windows. The approach supports multiple species views (activities, trace variants, directly-follows relations) and demonstrates improved accuracy and robustness to drift while maintaining reasonable latency and throughput. Empirical results on synthetic and real-world logs show favorable conformance metrics with linear scaling in the number of observed species, indicating practical applicability in real-time settings.

Abstract

Streaming process mining deals with the real-time analysis of event streams. A common approach for it is to adopt windowing mechanisms that select event data from a stream for subsequent analysis. However, the size of these windows denotes a crucial parameter, as it influences the representativeness of the window content and, by extension, of the analysis results. Given that process dynamics are subject to changes and potential concept drift, a static, fixed window size leads to inaccurate representations that introduce bias in the analysis. In this work, we present a novel approach for streaming process mining that addresses these limitations by adjusting window sizes. Specifically, we dynamically determine suitable window sizes based on estimators for the representativeness of samples as developed for species estimation in biodiversity research. Evaluation results on real-world data sets show improvements over existing approaches that adopt static window sizes in terms of accuracy and robustness to concept drifts.
Paper Structure (18 sections, 4 equations, 3 figures, 2 tables, 1 algorithm)

This paper contains 18 sections, 4 equations, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. Event Stream Model
  4. Windowing Methods
  5. Concept Drifts
  6. Completeness and Coverage Estimation
  7. Species Definitions
  8. Dynamic Window Size Selection
  9. Motivation for Window Representativeness
  10. Selecting a Window Size
  11. Estimations for Streaming Process Discovery
  12. Evaluation Results
  13. Experimental Setup
  14. Robustness to Concept Drifts
  15. Species Evaluation and Accuracy
  16. ...and 3 more sections

Figures (3)

  • Figure 1: The x-axis shows all windows found for the dataset, while the y-axis shows the number of events captured in the specific window. This demonstrates the adaption of the window to drifts.
  • Figure 2: Comparison of best/worst species with F1-scores
  • Figure 3: Analysis of the latency time for incrementally larger windows ($n_{\text{max}} = 500$)