An analysis of gamma-ray data collected at traffic intersections in Northern Virginia

TL;DR

This study collected and analyzed gamma-ray spectral data from traffic intersections in Northern Virginia over 15 months, using machine learning to identify anomalies and classify radioisotope patterns, with data made publicly available.

Contribution

It introduces a novel dataset of traffic-based gamma-ray spectra and applies machine learning for anomaly detection and isotope classification in this context.

Findings

Identification of radioisotope classes and their frequency patterns

Successful training of machine learning models on anomalous events

Data archive established for future research

Abstract

Gamma-ray spectral data were collected from sensors mounted to traffic signals around Northern Virginia. The data were collected over a span of approximately fifteen months. A subset of the data were analyzed manually and subsequently used to train machine-learning models to facilitate the evaluation of the remaining 50k anomalous events identified in the dataset. We describe the analysis approach used here and discuss the results in terms of radioisotope classes and frequency patterns over day-of-week and time-of-day spans. Data from this work has been archived and is available for future and ongoing research applications.