New insights into time series analysis
II - Non-correlated observations
1 SUPA (Scottish Universities Physics Alliance) Wide-Field Astronomy Unit, Institute for Astronomy, School of Physics and Astronomy, University of Edinburgh, Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK
2 Departamento de Física, Universidade Federal do Rio Grande do Norte, RN, 59072-970 Natal, Brazil
3 National Institute For Space Research (INPE/MCTI), Av. dos Astronautas, 1758 – SP, 12227-010 São José dos Campos, Brazil
Received: 21 November 2016
Accepted: 4 June 2017
Context. Statistical parameters are used to draw conclusions in a vast number of fields such as finance, weather, industrial, and science. These parameters are also used to identify variability patterns on photometric data to select non-stochastic variations that are indicative of astrophysical effects. New, more efficient, selection methods are mandatory to analyze the huge amount of astronomical data.
Aims. We seek to improve the current methods used to select non-stochastic variations on non-correlated data.
Methods. We used standard and new data-mining parameters to analyze non-correlated data to find the best way to discriminate between stochastic and non-stochastic variations. A new approach that includes a modified Strateva function was performed to select non-stochastic variations. Monte Carlo simulations and public time-domain data were used to estimate its accuracy and performance.
Results. We introduce 16 modified statistical parameters covering different features of statistical distribution such as average, dispersion, and shape parameters. Many dispersion and shape parameters are unbound parameters, i.e. equations that do not require the calculation of average. Unbound parameters are computed with single loop and hence decreasing running time. Moreover, the majority of these parameters have lower errors than previous parameters, which is mainly observed for distributions with few measurements. A set of non-correlated variability indices, sample size corrections, and a new noise model along with tests of different apertures and cut-offs on the data (BAS approach) are introduced. The number of mis-selections are reduced by about 520% using a single waveband and 1200% combining all wavebands. On the other hand, the even-mean also improves the correlated indices introduced in Paper I. The mis-selection rate is reduced by about 18% if the even-mean is used instead of the mean to compute the correlated indices in the WFCAM database. Even-statistics allows us to improve the effectiveness of both correlated and non-correlated indices.
Conclusions. The selection of non-stochastic variations is improved by non-correlated indices. The even-averages provide a better estimation of mean and median for almost all statistical distributions analyzed. The correlated variability indices, which are proposed in the first paper of this series, are also improved if the even-mean is used. The even-parameters will also be useful for classifying light curves in the last step of this project. We consider that the first step of this project, where we set new techniques and methods that provide a huge improvement on the efficiency of selection of variable stars, is now complete. Many of these techniques may be useful for a large number of fields. Next, we will commence a new step of this project regarding the analysis of period search methods.
Key words: methods: data analysis / methods: statistical / techniques: photometric / astronomical databases: miscellaneous / stars: variables: general / infrared: general
© ESO, 2017