| Abstract: | Radiological dirty bombs and improvised nuclear devices pose a significant threat to both public health and national security. Growing networks of radiation sensors have been deployed by a number of offices within the U.S. and international agencies. Detecting such threats while minimizing false alarm rates presents a considerable challenge to homeland security and public health. This research aims to achieve a higher probability of detection with a lower probability of false alarms. It focuses on the local spatial instability of radiation levels in order to detect radioactive materials based on robust outlier detection methods. Our approach includes a three‐step abnormality detection method consisting of one‐dimensional robust outlier detection for all gamma‐ray counts, a density‐based clustering analysis, and a two‐dimensional robust outlier detection method using a bagplot, based on spatial associations. The effectiveness of the method proposed is demonstrated through a case study, wherein radioactive materials are detected in urban environments, and its performance is compared with alternative methods employing a k‐sigma approach, local Getis–Ord ( ) statistic, and the goodness of fit of the Poisson distribution. |
