Home | Publications | RST+26

Efficient Event Sequence Predictions by Leveraging Event Context Clustering

MCML Authors

Simon Rauch

→ Group Thomas Seidl
Database Systems, Data Mining and AI

Gabriel Marques Tavares

Dr.

→ Group Thomas Seidl
Database Systems, Data Mining and AI

Thomas Seidl

Prof. Dr.

Director

Database Systems, Data Mining and AI

Abstract

Data-driven methods from the field of Predictive Process Monitoring (PPM) are able to perform different prominent forecasting and inference tasks, e.g., activity suffix prediction, remaining time prediction, and outcome prediction. Currently established deep learning frameworks in PPM offer the possibility to perform one specific or also multiple of these tasks inside a single predictive model. However, due to their model complexity, they suffer from high computational loads and lengthy training times, especially when the goal is to fit all possible downstream tasks into one holistic model. In contrast to that, we provide a time-efficient framework that does not rely on deep learning but nevertheless allows for the execution of any downstream task from the PPM domain while keeping the computational load reasonably small and, additionally, offering the potential for explainable results via rule extraction. By pairing event clustering on available intra-case context information with a suffix prediction model based on sequential patterns, we enable forecasting of any information associated with process events including predictions of temporal information and additional event attributes. Compared to deep learning techniques, we achieve similar or better results on common PPM tasks of suffix prediction and next timestamp prediction while performing the model training in a fraction of the time deep learning models need to be sufficiently trained.

inproceedings RST+26