Data Augmentation techniques in time series domain: a survey and taxonomy

• DOI: 10.1007/s00521-023-08459-3
• Publication Date: 2023-03-24T07:02:56Z
• Authors: Guillermo Iglesias; Edgar Talavera; Ángel González-Pr...; Alberto Mozo; Sandra Gómez-Canaval
• Keywords: Informática; FOS: Computer and information sciences; Computer Science - Machine Learning; 03 medical and health sciences; 0302 clinical medicine; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; I.2.6; Machine Learning (cs.LG)

Abstract With the latest advances in deep learning-based generative models, it has not taken long to take advantage of their remarkable performance area time series. Deep neural networks used work with series heavily depend on size and consistency datasets training. These features are usually abundant real world, where they limited often have constraints that must be guaranteed. Therefore, an effective way increase amount data is by using augmentation techniques, either adding noise or permutations generating new synthetic data. This systematically reviews current state art provide overview all available algorithms proposes a taxonomy most relevant research. The efficiency different variants will evaluated as central part process, well metrics evaluate main problems concerning each model analysed. ultimate aim this study summary evolution areas produce better results guide future researchers field.