Big Data Processing by Means of Unconventional Algorithm

Abstract

During last years, several successful algorithms emerged for classification of time series from various areas. But astronomical time series (a.k.a. light curves) are a bit more challenging to classify. They greatly vary in lengths, periods, noisiness and do not have clear borders between classes. In this work, we depict these issues on three publicly available data sets and several well-performing algorithms. We also present our approach that includes the use of artificial neural networks enhanced by evolutionary algorithm to find the best performing classification algorithm for pre-processed light curves with extracted features. Our approach is able to challenge results of related work that includes these data sets.