Enhancing operator performance during reactor design-based accidents, particularly those that could lead to the release of radioactive material, is critical for maintaining the safe operation of nuclear power plants (NPP) and ensuring the protection of human life and the environment. In this sense, integrating artificial intelligence (AI) into NPP safety systems offers fast and accurate fault predictions, enabling operators to make timely and informed decisions, thereby enhancing their reliability. However, operators need to remain “in the loop” to understand the model’s decision-making process, which is essential for establishing trust in the AI outcomes.

In this work, radionuclide concentration data are used to train several supervised models for accident classification, followed by the application of hyperparameter tuning techniques to optimize model performance. Finally, explainable artificial intelligence (XAI) is employed to uncover the reasoning behind the model predictions. The results showed that the random forest model demonstrated high performance, with strong accuracy and lower computational time. Additionally, the reasons for the model’s misclassifications are explained using XAI and validated in accordance with established engineering principles.