Messages of Mira variables in LMC & SMC with machine learning.

JiaYu Ou^1*, Chow-Choong Ngeow¹

¹Institute of Astronomy, National Central University, Taoyuan City, Taiwan

* Presenter:JiaYu Ou, email:m1039004@gm.astro.ncu.edu.tw

Nowadays , there are a lot of sky survey datasets with cadence of days, and machine learning is one of the most popular techniques to analyze the data due to its very powerful classification ability. There are several survey projects that we can obtain light curve data for interested variable stars, one of them being Mira variables.
Mira variables are asymptotic giant branch pulsating stars that exhibit large cyclical variation spanning 100 to 700 days, but in some extreme cases the variations can go beyond 1500 days.
Mira variables can be divided into O-rich and C-rich Miras with spectra, they can also be divided into Mira ,symbiotic Mira ,Mira with long term trend and long secondary periodic variables with some variations in their light curve. Our purpose is to use the machine learning technique for classifying various sub-classes of Mira. We collected 2015 confirmed Miras light curve data in LMC and SMC from OGLE database. Based on the light curves the Mira were divided into regular Miras and multi-periodic Miras.
We used python package Feature analysis for time series (FATs) to extract the light curve features, then we used these features to separate out the regular Mira and multi-periodic Mira using machine learning techniques. We found that in regular Miras the magnitude of maximum light can improve the period-luminosity relation, and we also found that regular Miras and multi-period Miras exhibit differences in color index using the OGLE photometric dataset.
Finally, we have collected SED data of regular Mira and multi-period Mira based on the SIMBAD database. We fit the SED component separated into the star part and dust shell part with blackbody radiation function. We found the multi-periodic Mira has much more dust component than regular Mira.

Keywords: AGB stars, variables