A Machine Learning Approach for Background Radiation Modeling and Anomaly Detection in Radiation Time Series pertained to Nuclear Security

摘要

In this work a new method is presented for background modeling and subsequent anomaly detection in radiation time series data. The underlying idea is the use of multiple identical learning Gaussian processes that are trained on data of various data lengths. The above idea allows the capturing of anomalies that occur while measurement are taken and are captured partially in some of the measured values.