ExplainabilitySafetyTrustworthinesscs.LG
Explainable AI to Improve Machine Learning Reliability for Industrial Cyber-Physical Systems
Annemarie Jutte, Uraz Odyurt
January 22, 2026
International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems
1 citations
Key Takeaways
-
XAI helps identify hidden training problems: By using explainability techniques (SHAP values), researchers discovered that ML models for industrial systems were making predictions based on insufficient context. This finding would have been invisible without opening up the "black box" of how the model reasons.
-
Better explanations lead to better performance: When the research team used XAI insights to adjust how much historical data the model could see (increasing the "window size"), predictive performance improved. This demonstrates that understanding why a model fails is the first step to fixing it.
-
Industrial systems demand transparency: Because industrial cyber-physical systems control critical infrastructure with safety and economic consequences, the paper emphasizes that ML models cannot simply be deployed as unexplainable black boxes—their reasoning must be auditable and understandable to human operators.
-
XAI is a practical debugging tool, not just theory: Rather than being an abstract concept, explainable AI functioned here as a concrete diagnostic method to reveal what information the model was actually using and what it was missing, enabling targeted improvements to training data.
Abstract
Industrial Cyber-Physical Systems (CPS) are sensitive infrastructure from both safety and economics perspectives, making their reliability critically important. Machine Learning (ML), specifically deep learning, is increasingly integrated in industrial CPS, but the inherent complexity of ML models results in non-transparent operation. Rigorous evaluation is needed to prevent models from exhibiting unexpected behaviour on future, unseen data. Explainable AI (XAI) can be used to uncover model reasoning, allowing a more extensive analysis of behaviour. We apply XAI to improve predictive performance of ML models intended for an industrial CPS use-case. We analyse the effects of components from time-series data decomposition on model predictions using SHAP values. Through this method, we observe evidence on the lack of sufficient contextual information during model training. By increasing the window size of data instances, informed by the XAI findings for this use-case, we are able to improve model performance.