Machine learning for radioxenon event classification for the Comprehensive Nuclear-Test-Ban Treaty.

Area of Science:

• Nuclear physics and atmospheric monitoring
• Machine learning applications in treaty verification

Background:

• The Comprehensive nuclear-Test-Ban-Treaty (CTBT) relies on detecting nuclear weapon tests.
• Radioxenon isotope monitoring in the atmosphere is a key detection method.
• Limited real-world data for nuclear explosion events necessitates simulation.

Purpose of the Study:

• To establish an effective classification model for nuclear explosion detection using machine learning.
• To evaluate the performance of various machine learning algorithms in analyzing radioxenon data.
• To assess the efficacy of these methods in high radioxenon background conditions.

Main Methods:

• Simulated nuclear explosion data sets based on real-world radioxenon measurements.
• Utilized radionuclide monitoring to measure activity concentrations of Xe-131m, Xe-133, Xe-133m, and Xe-135.
• Applied classic machine learning induction algorithms: Naïve Bayes, Neural Networks, Decision Trees, k-Nearest Neighbors, and Support Vector Machines.

Main Results:

• Machine learning algorithms successfully classified simulated nuclear explosion events.
• All tested induction algorithms demonstrated capability for this practical application.
• Machine learning methods significantly outperformed a simple linear discriminator in high radioxenon background environments.

Conclusions:

• Machine learning provides a robust framework for nuclear explosion detection via radioxenon analysis.
• These advanced algorithms are crucial for enhancing CTBT verification capabilities.
• The developed models offer improved detection accuracy, particularly under challenging environmental conditions.