A novel method based on data visual autoencoding for time series similarity matching

A variety of techniques are currently presented for querying and mining time series data based on different kinds of representations and similarity measures. These techniques mainly focus on the numerical characteristics of data and are sensitive to the changes of time series. However, we find that time series data generally contain curves sharing some set of visual characteristics and features. These characteristics offer a deeper understanding of time series data, and open up a potential new technique for time series analysis. Particularly beneficial from recent advances in deep neural networks (DNNs), representations and features can be automatically learnt by deep learning architectures such as autoencoders. In this paper, we propose a novel method, named Time Series Visualization (TSV), to efficiently match similar time series data. Architecture and algorithms of TSV based on stacked autoencoders are introduced in this paper. Further, important factors affecting the performance of TSV are discussed based on the empirical results. Through extensive empirical evaluation, it is demonstrated that TSV has a significant superiority in efficiency and accuracy compared to existing methods for time series similarity matching.