PuSH - Publication Server of Helmholtz Zentrum München: Efficient training of recurrent neural networks for remaining time prediction in predictive process monitoring.

Information

Clues to quality of journals

Open Access Policy of Helmholtz Association 2016

CC Licencences

Publication Metrics

Navigation

Home

Deutsch

EVA: Electronic Publishing

New EVA Application

Research

Advanced Search

Browse by ...

... HMGU-Authors/Consortia

... Organizational Structure

... Journal

... Publication Type

... Research Data

... Arbeitsgruppen

... Publication Year

Publication overview

Statistics

Statistics (last 5 years)

OA Publications

Publish

Submit Publication

Import publication from...

...EVA

Report missing publication

Highlights

Support & Contact

Contact persons

Help

Data protection

Helmholtz Open Science

JANE Journal Estimator

SHERPA/RoMEO

DOAJ

Export:

Text

Endnote (RIS) BIB

BibTeX

Roider, J.* ; Zanca, D.* ; Eskofier, B.M.

Efficient training of recurrent neural networks for remaining time prediction in predictive process monitoring.

In: (Business Process Management). Berlin [u.a.]: Springer, 2024. 238-255 (Lect. Notes Comput. Sc. ; 14940 LNCS)

DOI

Open Access Green as soon as Postprint is submitted to ZB.

Abstract
Metrics
Extra information

An important task in predictive process monitoring is the prediction of the remaining time. The accuracy of methods to solve this task has been steadily improved and verified during the last years. The current state-of-the-art uses Long Short-Term Memory (LSTM) neural networks, represented by the DA-LSTM model. However, training such methods requires substantial amounts of time and memory. This paper addresses specifically these problems. We adjust DA-LSTM and introduce DA-LSTM+ which achieves competitive error levels while reducing the training time significantly. Furthermore, we introduce trace-based sequence encoding as an alternative to prefix encoding, and an approach to use case attributes more efficiently to address time and memory limitations during training. The usage of these is not limited to LSTM’s but they are compatible with any neural network type. We evaluate them together with two alternatives for categorical feature encoding in an extensive benchmark study including eight different model architectures based on DA-LSTM+ and 14 publicly available datasets. The study shows that the training of neural network-based methods can be significantly accelerated with our contributions without affecting model’s performance. Our implementation is memory-efficient and publicly available.

Altmetric