A Framework for Object-Centric Predictive Monitoring of Collaborative Processes

Daniel Calegari; Andrea Delgado; Leonel Peña; Martín Rubio

cs.AIAug 27, 2026

A Framework for Object-Centric Predictive Monitoring of Collaborative Processes

Authors: Daniel Calegari, Andrea Delgado, Leonel Peña, Martín Rubio

Organizations: Universidad ORT Uruguay, Cuareim 1451, Montevideo, 11100, Montevideo, Uruguay · Instituto de Computación, Facultad de Ingeniería, Universidad de la República, J. Herrera y Reissig 565, Montevideo, 11300, Montevideo, Uruguay

Abstract

Predictive Process Monitoring (PPM) of collaborative, inter-organizational processes requires reasoning over multiple interdependent entities, including participants, messages, local executions, and the global collaboration case. Existing approaches extend traditional event logs with collaboration attributes but retain a single-case perspective, leaving much of this structure implicit. Object-centric process mining (OCPM) provides an alternative by representing these entities as first-class objects with explicit relations and multiple notions of case. This study connects collaborative PPM and OCPM through three contributions: (i) a formal semantic mapping from extended collaborative event logs to an OCED-conformant object-centric representation, serialized in OCEL 2.0; (ii) a reformulation of collaborative prediction tasks as object-centric prediction tasks; and (iii) a reproducible converter and prediction pipeline implementing the proposed mapping. We evaluate the framework on four public collaborative event logs and a fifth derived from the BPI Challenge 2013 incident-management log by executing the fourteen reformulated tasks using five predictive strategies across tabular, sequential, and graph-native encodings. We further discuss the benefits, limitations, and threats to the approach's validity. The representation makes collaboration structure explicit and makes it natural to state prediction targets based on object relations that fall outside the case-centric taxonomy, at the cost of increased relational complexity and dependence on object-centric tooling.

Figures & tables

Figure 1: A BPMN 2.0 healthcare business process collaboration from [ 7 ]

case	timestamp	activity	elemType	participant
case_459	2021-06-17 07:17:07	Communicate disease	SendTask	Patient
case_459	2021-06-17 07:17:10	Receive disease info	ReceiveTask	Gynecologist
case_459	2021-06-17 07:17:13	Examine patient	task	Gynecologist
case_459	2021-06-17 07:17:18	Blood draw	task	Gynecologist
case_459	2021-06-17 07:17:21	Send blood sample	SendTask	Gynecologist
case_459	2021-06-17 07:17:25	Receive blood sample	ReceiveTask	Laboratory

Table 1: Relevant data from the collaborative event log in the example case case_459 .

Category	Type Id	Prediction
Next Event	NE-NEPa	Next event that is likely to occur in a participant.
	NE-NEPr	Next event that is likely to occur in the process.
	NE-NPaA	Next participant that is likely to act.
	NE-NPaM	Next participant that is likely to send/receive a message.
	NE-NMPa	Next message that is likely to occur in a participant.
	NE-NMPr	Next message that is likely to occur in the process.

Table 2: Prediction types for collaborative processes from [ 10 ] .

Approach	Ref	Encoding	Model	Prediction types
(Adams, 2022)	[ 2 ]	Tabular, sequential, and graph	Regression, LSTM, and GNN	Remaining time
(Gherissi, 2023)	[ 17 ]	Flattening for a selected object type	LSTM (PPM-OC)	Next activity, next event time, remaining time
(Galanti, 2023)	[ 14 ]	Tabular over object interactions	Feature-based predictive models	KPI and outcome
(Adams, 2023)	[ 4 ]	Native graph	GNN, compared against graph embeddings and flattening	Next activity, next timestamp, remaining time
(Galanti, 2024)	[ 15 ]	Tabular, sequential, and graph	CatBoost, LSTM, and GNN; the GNN does not outperform	KPI
(Smit, 2024)	[ 23 ]	Native heterogeneous object-event graph (HOEG)	Heterogeneous GNN	Remaining time

Table 3: Object-centric PPM approaches: encoding, model, and prediction targets

Panel of Figure 3 — Figure 3: Instance-level representation of the first three events of collaboration case case_459 (Table 1 ). Dashed arrows represent qualified O2O relations, whereas solid arrows represent qualified E2O relations. Some attributes are omitted for readability.

Figure 4: High level pipeline of the framework.

Encoding	Input generation strategy	ML/DL models	Libraries
Tabular	Per-event vector, reading no object attribute: elapsed time; the number of distinct OrchestrationCase and Message objects up to the cut; and one column per activity, counting its occurrences among the cut event’s immediate predecessors.	Random Forest, XGBoost	scikit-learn
Sequential	The same flat feature vectors as the tabular encoding, stacked in prefix order into per-case sequences.	Transformers, LSTMs	PyTorch
Graph	Induced subgraphs of the OCPA event graph: the cut event plus its up to eight closest structural ancestor events, as interconnected nodes.	Graph Neural Networks (GNNs)	PyTorch

Table 5: Encoding approaches and compatible prediction models and libraries.

Statistic	Healthcare	Artificial 1	Artificial 5	Real 4	BPIC 2013
Cases	100	100	100	100	7,554
Events	1,450	800	2,360	1,800	69,584
Distinct activities	21	8	28	19	4
Max. events per case	18	8	25	18	135
Avg. events per case	14.5	8.0	23.6	18.0	9.21
Message types	8	2	6	4	1

Table 6: Descriptive statistics of the event logs.

Task	Target anchor	Problem type
NE-NPaA	Participant	Multiclass classification
NE-NMPr	Message	Multiclass classification
NV-PrT	CollaborationCase	Regression (time)
NV-PaT	OrchestrationCase	Regression (time)
NV-NMPr	Message	Count regression
OB-M	Message	Binary classification

Table 7: Representative subset for the end-to-end demonstration (RQ3).

	PredictCollab logs (78,600 fitted prefixes)				BPIC 2013 (653,298 fitted prefixes)
Predictor	Fit (s)	s/1k	Predict (ms)	Peak (MB)	Fit (s)	s/1k	Predict (ms)	Peak (MB)
Random Forest	41	0.10	30.2	899	84	0.03	104.5	6,411
XGBoost	62	0.16	4.4	964	103	0.03	44.6	2,024
Transformer	886	2.25	3.1	1,034	13,513	4.14	282.8	2,192
LSTM	292	0.74	1.2	962	10,753	3.29	26.1	1,997
GNN	5,389	13.71	204.3	1,074	43,994	13.46	6,837.9	2,748

Table 9: RQ3 execution cost over the full 14-task catalog, by predictor and log group.

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Task	Parameter	Output / decoder	Accessors used	Lemma	$\Theta_{\tau}^{\mathit{src}}$
NE-NEPr	—	$\mathcal{A}_{\mu}/\mathrm{dec}_{\mathcal{A}}$	$\mathit{evtype}$	(i)	[ 10 ] , Def. 3
NE-NPaA	—	$\mathsf{Pa}_{\mu}/\mathrm{dec}_{\mathsf{Pa}}$	$\mathit{pa}$	(ii)	[ 10 ] , Def. 4
NE-NEPa †	—	$\mathcal{A}_{\mu}\!\times\!\mathsf{Pa}_{\mu}/(\mathrm{dec}_{\mathcal{A}},\mathrm{dec}_{\mathsf{Pa}})$	$\mathit{evtype}$ , $\mathit{pa}$	(i), (ii)	[ 10 ] , Def. 5
NE-NPaM †	$d\in\{\mathit{snd},\mathit{rcv}\}$	$\mathsf{Pa}_{\mu}^{\bot}/\mathrm{dec}_{\mathsf{Pa}}^{\bot}$	$\mathit{msg}$ , from / to	(iii)	[ 10 ] , Def. 6
NE-NMPr	—	$\mathcal{A}_{\mu}^{\bot}/\mathrm{dec}_{\mathcal{A}}^{\bot}$	$\mathit{isMsg}$ , $\mathit{evtype}$	(i), (iii)	Table 2 only (above)
NE-NMPa	$p\in\mathcal{P}_{L}$	$\mathcal{A}_{\mu}^{\bot}/\mathrm{dec}_{\mathcal{A}}^{\bot}$	$\mathit{isMsg}$ , $\mathit{oc}$ , $\mathit{evtype}$	(i)–(iii)	Table 2 only (above)

Appendix

Table 10: Parameters, output typing, accessors, and source-level counterparts of the tasks in Proposition 2 . Types marked

\dagger

are documented adaptations, not literal restatements, of the cited definition (see "Departures from a literal restatement" below).

Process Mining

Process mining turns event logs of real-world activities into models and predictions for understanding, checking, and improving workflows.

See all 45 papers

On the Potential of Multi-Task Learning in Predictive Process Monitoring
Sep 11, 2026Lukas Kirchdorfer, Keyvan Amiri Elyasi, Heiner StuckenschmidtPredictive Process Monitoring Multi-Task Learning
University of Mannheim
GNN4PPM: Multi-Target Predictive Process Monitoring with Relational Graph Convolutional Networks
Sep 13, 2026Ana Costa, Johannes Mäkelburg, Luise PufahlPredictive Process Monitoring Process Mining
Technical University of Munich, School of Computation, Information and Technology, Heilbronn, Germany
EHHN: An Event-driven Heterogeneous Hypergraph Network for Object-Centric Next Activity Prediction
Jul 2, 2026Jiaxing Wang, Kaitao Chen, Zhubin Han +4Temporal GNNs Process Mining
College of Computer Science and Technology, Zhejiang University of Technology, 310023, Hangzhou, China · University of Southern Queensland, Toowoomba, QLD 4350, Australia

Sep 11, 2026J/K move · Enter open · S save

On the Potential of Multi-Task Learning in Predictive Process Monitoring

Lukas Kirchdorfer, Keyvan Amiri Elyasi, Heiner Stuckenschmidt

University of Mannheim

Predictive Process Monitoring Multi-Task Learning Multi-Task Optimization

Open paper AskPDF arXiv

Predictive Process Monitoring (PPM) forecasts how ongoing organizational processes unfold, enabling information systems to move beyond execution support toward proactive analysis and monitoring. Although deep learning has improved prediction accuracy in PPM, most approaches follow a single-task learning (STL) setup, training a separate model per task. This increases maintenance effort and overlooks potential synergies. Multi-task learning (MTL), which jointly learns multiple prediction targets in one model, offers a promising alternative, yet its effectiveness in PPM remains underexplored. It remains unclear whether and under which settings MTL improves upon STL, which prediction tasks benefit most from joint learning, which task combinations are particularly synergistic, and if and how tasks should be balanced. To fill this gap, we present the first comprehensive empirical study of MTL for PPM, evaluating a variety of task combinations, neural architectures, and optimization methods. Overall, our results position MTL as a strong paradigm for PPM: we see substantial improvements in next-activity prediction and inherent mitigation of class imbalance using MTL, while task balancing is especially critical under low-capacity models.

Abstract

Process Mining

Explore similar work

On the Potential of Multi-Task Learning in Predictive Process Monitoring

GNN4PPM: Multi-Target Predictive Process Monitoring with Relational Graph Convolutional Networks

EHHN: An Event-driven Heterogeneous Hypergraph Network for Object-Centric Next Activity Prediction