Radiological Clearance at CERN: Classical Machine Learning Methods for Waste Classification

At CERN, the maintenance and dismantling of high-energy particle accelerators generate waste that is potentially radioactive. When activity levels are negligible, waste can be released from regulatory control through clearance procedures. However, radiological characterization is challenging due to the diversity of activation scenarios, radionuclide inventories, and material compositions. We address this complexity by developing a data-driven probabilistic approach that combines activation simulations with classical machine learning. This study applies the method to irradiated electric cables, with a particular focus on defining threshold values for measurable quantities (activity, dose rate, and mass) that determine clearance eligibility. The proposed method offers operational advantages by simplifying measurement and decision-making procedures and is consistent with international guidance on clearance [1, 2].