We are pleased to announce that our paper “Irreconcilable Differences? Investigating Consensus of Post-hoc XAI for ML-NIDS via Decomposition” was awarded the Best Student Paper Award at the 21st International Conference on Network and Service Management (CNSM 2025), which took place from October 27 to 31, 2025, in Bologna, Italy.

The paper is the result of a collaboration between researchers at the University of Würzburg and our department. Katharina Dietz presented the joint work at the conference and received the Best Student Paper Award.

In this work, we investigate how feature correlations affect the consistency of explanations in Explainable Artificial Intelligence (XAI) for machine learning-based network intrusion detection systems (ML-NIDS). We focus on post-hoc XAI methods, which often provide conflicting explanations due to correlated features, and explore whether feature decorrelation via decomposition techniques can improve agreement among them. 
Overall, we find that feature decorrelation can significantly increase consensus among XAI methods, but its effectiveness depends strongly on the model, preprocessing, and dataset characteristics. Sparse PCA can recover some interpretability, though results vary with the level of sparsity enforced.

These findings offer practical insights for enhancing the reliability and trustworthiness of XAI explanations in network security applications.
Paper: Katharina Dietz, Johannes Schleicher, Stefan Geissler, Michael Seufert, Tobias Hoßfeld. "Irreconcible Differences? Investigating Consensus of Post-hoc XAI for ML-NIDS via Decomposition." 21st International Conference on Network and Service Management (CNSM), October 27-October 31, 2025, Bologna, Italy.

Link to paper:  Irreconcilable Differences? Investigating Consensus of Post-hoc XAI for ML-NIDS via Decomposition