Machine Learning for Transient Recognition in Difference Imaging With Minimum Sampling Effort

TL;DR

This paper introduces a minimal sampling effort approach for creating training sets for machine learning classifiers to distinguish real astronomical transients from bogus detections, significantly reducing manual labeling.

Contribution

It proposes a novel data labeling strategy using all detections in science and difference images, enabling effective training with minimal human effort.

Findings

Achieves up to 95% accuracy in real-bogus classification

Maintains a false alarm rate of 1%

Reduces manual labeling effort in training data creation

Abstract

The amount of observational data produced by time-domain astronomy is exponentially in-creasing. Human inspection alone is not an effective way to identify genuine transients fromthe data. An automatic real-bogus classifier is needed and machine learning techniques are commonly used to achieve this goal. Building a training set with a sufficiently large number of verified transients is challenging, due to the requirement of human verification. We presentan approach for creating a training set by using all detections in the science images to be thesample of real detections and all detections in the difference images, which are generated by the process of difference imaging to detect transients, to be the samples of bogus detections. This strategy effectively minimizes the labour involved in the data labelling for supervised machine learning methods. We demonstrate the utility of the training set by using it to train several classifiers utilizing as the feature representation the normalized pixel values in 21-by-21pixel stamps centered at the detection position, observed with the Gravitational-wave Optical Transient Observer (GOTO) prototype. The real-bogus classifier trained with this strategy can provide up to 95% prediction accuracy on the real detections at a false alarm rate of 1%.