|
A simple and calibrated approach for uncertainty-aware remaining time prediction
Amiri Elyasi, Keyvan
;
van der Aa, Han
;
Stuckenschmidt, Heiner
|
DOI:
|
https://doi.org/10.1007/978-3-032-02867-9_14
|
|
URL:
|
https://link.springer.com/chapter/10.1007/978-3-03...
|
|
Document Type:
|
Conference or workshop publication
|
|
Year of publication:
|
2025
|
|
Book title:
|
Business Process Management : 23rd International Conference, BPM 2025, Seville, Spain, August 31 - September 5, 2025
|
|
The title of a journal, publication series:
|
Lecture Notes in Computer Science
|
|
Volume:
|
16044
|
|
Page range:
|
217-234
|
|
Conference title:
|
23rd International Conference on Business Process Management (BPM 2025)
|
|
Location of the conference venue:
|
Sevilla, Spain
|
|
Date of the conference:
|
31.08.-05.09.2025
|
|
Publisher:
|
A. Reijers, Hajo
;
Cabanillas, Cristina
;
Senderovich, Arik
;
Vanderfeesten, Irene
|
|
Place of publication:
|
Berlin [u.a.]
|
|
Publishing house:
|
Springer
|
|
ISBN:
|
978-3-032-02866-2 , 978-3-032-02868-6 , 978-3-032-02867-9
|
|
ISSN:
|
0302-9743 , 1611-3349
|
|
Publication language:
|
English
|
|
Institution:
|
School of Business Informatics and Mathematics > Practical Computer Science II: Artificial Intelligence (Stuckenschmidt 2009-)
|
|
Subject:
|
004 Computer science, internet
|
|
Keywords (English):
|
remaining time prediction , uncertainty quantification , predictive process monitoring , deep learning
|
|
Abstract:
|
Business processes are typically supported by information systems that log execution data, enabling the prediction of remaining time for ongoing process instances. Deep learning models are often used for this task due to their accuracy, but they only provide point estimates, without accounting for uncertainty. This limits their reliability, as decision-making often benefits from prediction intervals. Uncertainty quantification techniques can help by estimating both expected values and uncertainty. However, existing techniques are often poorly calibrated, computationally expensive, or not adaptable to different deep learning models. This paper examines these challenges and proposes a simple, efficient solution using Laplace approximation and calibrated regression. Our approach distinguishes between model and data uncertainty, integrates easily with any deep learning model, and can be applied to pre-trained networks. Benchmarking on 10 real-world event logs shows that our method matches state-of-the-art performance while significantly reducing training and inference time. This makes it a strong yet simple baseline for uncertainty-aware remaining time prediction in business processes.
|
 | Dieser Eintrag ist Teil der Universitätsbibliographie. |
 Search Authors in
You have found an error? Please let us know about your desired correction here: E-Mail
Actions (login required)
 |
Show item |
|
