Auto-TSA: An Automatic Time Series Analysis System Based on Meta-learning

Time series is a necessary data type in both industrial scenarios and data analysis. In this era of explosive data growth, the significant development of sensors has made it possible to obtain massive amounts of time series data. However, the performance of different algorithms for different types of time series data varies greatly. So how to automatically choose an optimal algorithm for different data becomes the key to improving efficiency and saving resources. However, existing cloud services or open-source frameworks are not easy to use for users with little experience in time series analysis, and the user needs to rely on continuous experimentation to choose an algorithm that best suits his scenario. This significantly reduces the efficiency of data analysis. Thus to address this phenomenon, in this paper we propose an automated time series analysis system Auto-TSA. The biggest advantage over existing methods is that we have designed “Automation mode” so that even first-time users can easily use it. Users can automatically obtain the best-performing algorithm and hyperparameter configuration by entering only their own data. The whole process of automatic algorithm selection and hyperparameter optimization is efficient by introducing historical experience. In addition, we also design “Customization mode” for time series analysis experts, which makes it easier to use and more functional through the excellent and simple interface design. We will describe the framework and workflow of the system in detail, and show some samples to guide users to use it quickly.