When data science and physics-ignorant machine learning produces hallucinations: Comment on “Machine Learning-Driven Prediction of Gamma-Ray Shielding Performance in TeO2−Bi2O3−ZnO−BaF2 Glasses” [Tang, Y., Radiation Physics and Chemistry, https://doi.org/10.1016/j.radphyschem.2025.113586]
Autor(in); Institution
Rabus, Hans; 8, Medizinphysik und metrologische Informationstechnik, PTB-Berlin
Quelle/Jahr
Radiation Physics and Chemistry: 243
(2026), 1
- 5
The paper by Tang presents an automated approach to determining optimal machine-learning based regressions of mass-attenuation data. The conclusions of the paper are based on a deceptive omission of a comparison between model predictions and data used for training and testing which would have elucidated that such conclusions are factually wrong. The data-pre-processing used to optimize training is not suited for this purpose in the way it was applied. Most of the regression techniques reported as sub-optimal produce larger variance than the data which indicates that the model training failed. Therefore, the paper must be stated to contain methodological flaws and drawing factually incorrect conclusions.
Themenbereich der Metrologie
Metrologie in der Medizin
Zitierung
Rabus, H. (2026). When data science and physics-ignorant machine learning produces hallucinations: Comment on “Machine Learning-Driven Prediction of Gamma-Ray Shielding Performance in TeO2−Bi2O3−ZnO−BaF2 Glasses” [Tang, Y., Radiation Physics and Chemistry, https://doi.org/10.1016/j.radphyschem.2025.113586]. Radiation Physics and Chemistry, 243, 1–5. https://doi.org/10.1016/j.radphyschem.2026.113733
