Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HOEG: A New Approach for Object-Centric Predictive Process Monitoring

Tim K. Smit, Hajo A. Reijers, Xixi Lu

TL;DR

The paper tackles predictive process monitoring on object-centric event logs by introducing HOEG, a native heterogeneous object event graph that encodes events, multiple object types, and their interactions without aggregating attributes. HOEG enables learning via a heterogeneous GNN that traverses typed edges to integrate event and object information for remaining-time prediction. Empirical results across three OCELs show HOEG can outperform OCEL baselines like EFG when object attributes and interactions are informative, though it incurs higher training/prediction times and exhibits dataset-dependent performance. Overall, HOEG presents a promising, OCEL-native encoding approach with trade-offs between predictive performance and scalability, and future work includes richer edge features and multi-entity prediction tasks such as outlier detection.

Abstract

Predictive Process Monitoring focuses on predicting future states of ongoing process executions, such as forecasting the remaining time. Recent developments in Object-Centric Process Mining have enriched event data with objects and their explicit relations between events. To leverage this enriched data, we propose the Heterogeneous Object Event Graph encoding (HOEG), which integrates events and objects into a graph structure with diverse node types. It does so without aggregating object features, thus creating a more nuanced and informative representation. We then adopt a heterogeneous Graph Neural Network architecture, which incorporates these diverse object features in prediction tasks. We evaluate the performance and scalability of HOEG in predicting remaining time, benchmarking it against two established graph-based encodings and two baseline models. Our evaluation uses three Object-Centric Event Logs (OCELs), including one from a real-life process at a major Dutch financial institution. The results indicate that HOEG competes well with existing models and surpasses them when OCELs contain informative object attributes and event-object interactions.

HOEG: A New Approach for Object-Centric Predictive Process Monitoring

TL;DR

The paper tackles predictive process monitoring on object-centric event logs by introducing HOEG, a native heterogeneous object event graph that encodes events, multiple object types, and their interactions without aggregating attributes. HOEG enables learning via a heterogeneous GNN that traverses typed edges to integrate event and object information for remaining-time prediction. Empirical results across three OCELs show HOEG can outperform OCEL baselines like EFG when object attributes and interactions are informative, though it incurs higher training/prediction times and exhibits dataset-dependent performance. Overall, HOEG presents a promising, OCEL-native encoding approach with trade-offs between predictive performance and scalability, and future work includes richer edge features and multi-entity prediction tasks such as outlier detection.

Abstract

Predictive Process Monitoring focuses on predicting future states of ongoing process executions, such as forecasting the remaining time. Recent developments in Object-Centric Process Mining have enriched event data with objects and their explicit relations between events. To leverage this enriched data, we propose the Heterogeneous Object Event Graph encoding (HOEG), which integrates events and objects into a graph structure with diverse node types. It does so without aggregating object features, thus creating a more nuanced and informative representation. We then adopt a heterogeneous Graph Neural Network architecture, which incorporates these diverse object features in prediction tasks. We evaluate the performance and scalability of HOEG in predicting remaining time, benchmarking it against two established graph-based encodings and two baseline models. Our evaluation uses three Object-Centric Event Logs (OCELs), including one from a real-life process at a major Dutch financial institution. The results indicate that HOEG competes well with existing models and surpasses them when OCELs contain informative object attributes and event-object interactions.
Paper Structure (17 sections, 2 equations, 3 figures, 6 tables)

This paper contains 17 sections, 2 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. HOEG Approach
  5. Machine Learning on HOEG
  6. Evaluation
  7. Feature Encodings
  8. Models and Hyperparameters
  9. Baselines and Evaluation Metrics
  10. Object-Centric Datasets
  11. Results
  12. Hyperparameter Tuning Experiment
  13. Encoding Type Experiment
  14. Baseline Experiment
  15. Discussion
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Heterogeneous object event graph for Execution A (derived from Tables \ref{['tab:ocel_events_otc_example']} and \ref{['tab:ocel_objects_otc_example']}) of the running OTC example. Note: the trimmed (for readability) edges going out from Ship package (e9) are connected with objects o1, o2, i1, i3, p1, p2. Faded event node e10 signifies future event Confirm delivery. Also note that for readability, not all edges are drawn.
  • Figure 2: Test MAE scores for EFG and HOEG per hyperparameter setting. Note: scales (y-axis) are not aligned, as we intend to compare hyperparameter settings within each encoding and dataset.
  • Figure 3: Violin plot of the MAE score distribution over the hyperparameter settings per split (a, c, e). Training and validation loss learning curves of tuned models (b, d, f). BPI17: (a) and (b), OTC: (c) and (d), FI: (e) and (f).

Theorems & Definitions (5)

  • definition thmcounterdefinition: Event Log with Objects
  • definition thmcounterdefinition: Object Graph
  • definition thmcounterdefinition: Process Execution
  • definition thmcounterdefinition: Connected Component Extraction
  • definition thmcounterdefinition: HOEG