In an era where nuclear non-proliferation hangs in the balance, scientists have handed the UN’s atomic watchdog a potential game-changer: artificial intelligence that can assess uranium enrichment with almost flawless precision.
Researchers at the University of Sharjah say they have built an AI system that predicts uranium enrichment levels with extraordinary accuracy while simultaneously identifying different types of nuclear material — all from gamma spectroscopy data.
The study, published in the journal Annals of Nuclear Energy, put three machine-learning models — XGBoost, CatBoost, and LightGBM — through their paces. The results could hand the International Atomic Energy Agency (IAEA) powerful new tools to enforce the Nuclear Non-Proliferation Treaty (NPT).
Also worth a look: Nexdata’s 10,000-hour robot training dataset captures human motion from five angles and Florida law firm warns chatbot legal advice could sink injury claims.
Strikingly, all three models achieved perfect scores in identifying uranium categories. Each recorded 100% accuracy, precision, recall, and F1-score when classifying material as highly enriched uranium (HEU), low-enriched uranium (LEU), or natural uranium (NU).
The framework tackles two jobs at once: regression to predict the percentage of uranium-235 (235U) enrichment, and classification to identify the material type. The models were trained on structured datasets derived from IAEA benchmark simulations of high-purity germanium (HPGe) detector outputs — the industry standard for isotopic assay.
“The regression component achieved state-of-the-art performance, with XGBoost yielding a mean squared error (MSE) of 0.0019, a mean absolute error (MAE) of 0.0281, and a coefficient of determination (R²) of 1.0000,” the authors report. “LightGBM and CatBoost followed with R² values of 0.99999 and 0.9971, respectively.”
Among the trio, XGBoost delivered the strongest overall performance in predicting enrichment levels, with a near-negligible prediction error and a perfect R² score of 1.0000. LightGBM came close, achieving nearly identical accuracy while offering faster computation times. CatBoost, meanwhile, proved particularly robust with smaller datasets, maintaining stable and consistent results even with limited data.
The researchers maintain that the system could prove critical to improving nuclear safeguards by providing faster and more accurate analysis of nuclear materials, potentially boosting the ability of UN inspectors and regulators to monitor compliance with international non-proliferation agreements.
“These results confirm the robustness and generalizability of gradient-boosted decision tree algorithms in the highly sensitive domain of nuclear safeguards,” the authors write. “The proposed framework demonstrates the potential of machine learning as a decision-support tool for selected safeguards activities involving gamma spectroscopy.”
The methodology addresses two critical analytical tasks relevant to IAEA verification activities that involve non-destructive assay (NDA). As the authors note, inspectors may need to simultaneously determine quantitative enrichment levels and qualitative material classifications to support safeguards verification and anomaly detection.
To simulate real-world conditions, the study used structured gamma spectroscopy datasets representing measurements obtained using HPGe detectors. Each model was trained independently for regression and classification tasks using engineered and isotope-derived features, and evaluated across metrics including MSE, MAE, R², accuracy, precision, recall, and F1-score.
XGBoost emerged as the most accurate and generalizable model. Its fine-grained tree construction, combined with robust regularization and carefully optimized hyperparameters, appears to have enabled it to capture complex nonlinear relationships within the spectral and isotopic feature space. In safeguards applications involving gamma spectroscopy, such predictive accuracy could provide valuable decision support for enrichment estimation — subject to further validation under real-world inspection conditions.
CatBoost also delivered strong performance, achieving an MSE of 2.7757, MAE of 0.7580, and R² of 0.9971, though it exhibited slightly greater residual variance. The higher error, particularly at lower enrichment levels, may reflect the model’s sensitivity to data sparsity. Still, CatBoost attained perfect classification accuracy (100%) with precision, recall, and F1-scores of 1.00 across all material categories. Its symmetric tree architecture and stable gradient estimation enabled it to distinguish effectively between enrichment classes despite imbalances in class distribution.
LightGBM combined exceptional computational efficiency with high predictive accuracy. In the regression task, it achieved an MSE of 0.0119, MAE of 0.0389, and an R² score of 0.99999 — closely approaching XGBoost’s performance. Achieving such results while requiring substantially less training time and memory underscores LightGBM’s potential for low-latency safeguards applications.
Still, the researchers are careful not to overstate their findings. The scope of the study is limited to evaluating machine-learning algorithms using benchmark gamma spectroscopy datasets. The results have yet to be tested under the operational realities of IAEA safeguards, where verification activities are governed by safeguards agreements, facility-specific arrangements, inspection objectives, measurement uncertainty requirements, and a host of legal and operational considerations.
As a result, the findings should be interpreted as evidence of the technical potential of AI-assisted spectral analysis rather than as proof of enhanced safeguards effectiveness. The authors emphasize that future research should test these methods on representative operational datasets and in collaboration with safeguards stakeholders to better evaluate their suitability within existing verification frameworks.
Original Source URL: https://www.sciencedirect.com/science/article/abs/pii/S030645492600681X?via%3Dihub
Why it matters: This research suggests that machine learning could dramatically enhance the speed and accuracy of nuclear material verification — a critical need as geopolitical tensions rise. If validated under real inspection conditions, these AI tools could help the IAEA catch diversion of nuclear material faster and more reliably, strengthening the global non-proliferation regime.
LEON BARKHO
University Of Sharjah
+971 50 165 4376
