Process Model Forecasting
Processes change over time: some activities become more frequent, while others disappear. Process model forecasting predicts these changes across the whole process, helping us anticipate its future structure and behaviour.
Time Series Foundation Models for Process Model Forecasting
CAiSE 2026.
We study whether models pretrained on large collections of time series can forecast changes in process behaviour. We compare using these models directly with adapting them to process data.
A benchmarking study on process model forecasting: univariate vs. multivariate approaches
Process Science, 2025.
We compare methods that forecast activity-to-activity transitions separately with methods that model several transitions together. The benchmark covers statistical, machine-learning, and deep-learning approaches on real event logs.
Multivariate Approaches for Process Model Forecasting
ML4PM at ICPM 2024 · Proceedings published in 2025 · Best Paper Award.
This work investigates forecasting several process transitions jointly to capture relationships between them. It compares machine-learning and deep-learning methods across different real processes.