Revolutionizing nanosatellites’ data integrity with SEEnet: A real-time ensemble learning approach for Single-Event Effect (SEE) prediction

As nanosatellites make access to space more affordable and widespread, protecting onboard data from radiation-related damage has become a major challenge for modern low-cost missions. These small satellites often rely on commercial off-the-shelf (COTS) electronic components, which are particularly vulnerable to radiation-induced Single-Event Effects (SEEs) that can disrupt system operation and compromise mission data integrity. To address this challenge, our study proposes SEEnet, a lightweight ensemble learning framework designed for real-time prediction of SEE occurrence in nanosatellite systems. SEEnet combines multiple decision-tree classifiers with varying model depths using a soft-voting strategy, allowing it to improve prediction reliability while maintaining low computational complexity suitable for resource-constrained onboard environments. Our proposed approach is evaluated using a publicly available dataset from the Institute of Space Systems, University of Stuttgart, describing satellite spatial characteristics and associated SEE events. Experimental results show that SEEnet achieves a classification accuracy of 77%, outperforming several baseline machine-learning models, including Support Vector Machines, Random Forests, and Gradient Boosting, under the same evaluation conditions. In addition, the model demonstrates balanced precision-recall performance and provides bootstrap-based uncertainty estimates, enhancing confidence in its predictions. Overall, the results indicate that SEEnet offers an effective and computationally efficient solution for early SEE risk assessment, supporting proactive fault mitigation and improved data reliability in real-time nanosatellite missions.