Transient Detection in Light Curve Using Random Cut Forest

Abstract

Along with the improvement of temporal resolution in so-called "time-domain astronomy", it has become possible to observe celestial phenomena on second-scale. Discovering unknown second-scale phenomena is an unexplored frontier in astronomy. However, the amount of data to be handled is enormous, and the targets to be discovered are unknown. Therefore, such a data-driven detection methodology that can be used in a high-speed and memory-saving environment is required. In this study, we first create a benchmark dataset for the transient detection task based on the real light curve data, which include randomly generated transient patterns with specific shapes and degrees. Then, we apply an anomaly detection method called Random Cut Forest for detecting the artificial outbursts in the benchmark. The results are then compared with the performance of the previously proposed method based on probability inequalities.