Detection for Transient Patterns with Unpredictable Duration using Chebyshev Inequality and Dynamic Binning
Thanapol Phungtua-Eng, Yoshitaka Yamamoto, Shigeyuki Sako · 2021
This paper proposes an online algorithm for detecting transient patterns with an unpredictable duration from an astronomical data stream. The key idea appears in the binning technique which is often useful for noise filtering. It is used to split the original data stream into bins with a fixed length. A suitable length for binning depends on the transient patterns to be detected, although their prior information is not available in advance. The proposed method addresses this problem by introducing a dynamic mechanism which enables to adjust the bins with an adequate length. Consequently, it prevents from missing transient patterns with arbitrary duration Our extensive experiments using real astronomy dataset reveals that the proposed method outperforms the existing methods with high accuracy.