EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3282. Introducing SuTraN: A Transformer-Based Encoder-Decoder Model for Enhanced Suffix Prediction in Business Process Monitoring

Invited abstract in session TB-27: Advances in Process Mining, stream Mathematical Optimization for XAI.

Tuesday, 10:30-12:00
Room: 047 (building: 208)

Authors (first author is the speaker)

1. Brecht Wuyts
LIRIS, KU Leuven

Abstract

Predictive Process Monitoring (PPM) enhances the field of Process Mining (PM) by applying predictive analytics to forecast the future course of ongoing business processes. The prediction of suffixes, which involves predicting the sequence of forthcoming events, including their activity labels, timestamps, and remaining runtime, is particularly intricate. Present methodologies predominantly encounter two issues: firstly, they are predominantly trained to anticipate the next event, employing iterative feedback mechanisms for generating suffixes upon deployment—a strategy that not only introduces inaccuracies but also limits the exploitation of informative event attributes; secondly, the prevalent reliance on Long Short-Term Memory (LSTM) networks impedes their performance due to their inherent challenges with long-range dependencies and information loss over lengthy sequences. In response, this study unveils the Suffix Transformer Network (SuTraN), pioneering the use of an encoder-decoder Transformer framework for comprehensive suffix prediction in PPM. Tailored for this purpose, SuTraN innovatively forecasts complete event suffixes through a singular forward pass, eliminating the need for iterative loops and leveraging the full spectrum of event data without the constraints found in LSTM-dependent methods. Our evaluations, conducted on several real-world event logs, confirm that SuTraN delivers improved performance in suffix prediction relative to established methods.

Keywords

Status: accepted

Back to the list of papers