Comparing prominent periodicity detection algorithms with synthetic and real data to determine a superior method.

In the process of discovering new exoplanets, use of periodicity detection algorithms is a common method used to analyze transit data. The main issue with these algorithms when going through large tera/peta-scale datasets is that they can be computationally expensive and slow. In addition, when dealing with unevenly sampled time-series data with a low signal-to-noise ratio, it is challenging for periodicity detection algorithms to identify the true period. The Lomb-Scargle periodogram is a well-known least-squares method that tackles this issue, specializing in detecting periods within unevenly sampled series. Conditional entropy is another widely-used periodicity detection method that employs a correntropy-based partitioning scheme to represent data. Machine learning is also a newer method that has recently been used in time series analysis. The purpose of this study is to compare the accuracy and computational efficiency of these methods. With python, we create synthetic data and use AstroML’s database to experiment with real data to investigate which method is superior under different conditions by varying signal-to-noise ratios, sample sizes and types of noise.