Towards detection of anomalous cosmic ray signals for observations acquired from Cosmic Ray Extremely Distributed Observatory mobile detectors

In this work, we propose and test a method that allows the detection of anomalous cosmic ray signals acquired using Complementary Metal-Oxide-Semiconductor detectors. The method uses unsupervised embedding based on Principal Component Analysis which we named Eigenhits in apparent analogy to Eigenfaces. The embedding generated using Eigenhits allows the detection of potential anomalies, defined as images whose position described by the embedding relative to a given measure is above a certain distance threshold from other images. Thus, the problem of anomaly detection was reduced to the problem of detecting outliers which can be solved, for example, using clustering algorithms. We conducted tests of our approach on the Cosmic Ray Extremely Distributed Observatory (CREDO) dataset containing 13168 images and obtained satisfactory results demonstrating the stability and effectiveness of the method. The embedding generation method we propose in this paper and the evaluation of its effectiveness in detecting anomalies in images of cosmic ray events from CREDO dataset is a pioneering study with many critical applications.