XAI & Robustness in PPM

A useful prediction should come with an explanation people can understand and remain reliable when its inputs change. Together with Alexander Stevens and co-authors, we study explanation quality and test predictive models against deliberately altered process data.

Generating Realistic Adversarial Examples for Business Processes using Variational Autoencoders

Journal submission; preprint (2024).

We generate misleading examples by changing a learned representation of process executions, rather than directly editing their events. This helps keep the examples close to realistic process behaviour when testing model vulnerabilities.

Preprint

Manifold Learning for Adversarial Robustness in Predictive Process Monitoring

ICPM 2023.

We use manifold learning to keep adversarial examples close to the patterns found in real process data. Training on these examples explores how models can become more resistant to attacks while retaining predictive accuracy.

Paper

Assessing the Robustness in Predictive Process Monitoring through Adversarial Attacks

ICPM 2022.

We introduce a framework for testing how deliberately altered inputs affect predictions and their explanations. It compares predictive models and evaluates whether training with attacked data improves their robustness.

Paper · Code

Quantifying Explainability in Outcome-Oriented Predictive Process Monitoring

ML4PM / ICPM 2021; proceedings published in 2022.

We measure explanations by how understandable they are and how faithfully they describe a model’s behaviour. This allows us to compare inherently interpretable models with explanations added to more complex models.

Paper

Outcome-Oriented Predictive Process Monitoring on Positive and Unlabelled Event Logs

Related work with Alexander Stevens · ML4PM / ICPM 2022.

We also study prediction when only some positive outcomes are labelled, such as detected fraud. The approach learns from these incomplete labels without treating every unlabelled case as negative.

Topic page · Paper

All research topics