A time series contain a large amount of information suitable for forecasting. Classical statistical and recent deep learning models have been widely used in a variety of forecasting applications. During the training data preparation stage, most models collect samples by sliding a fixed-sized window over the time axis of the input time series. We refer to this conventional method as "sparse sampling" because it cannot extract sufficient samples because it ignores another important axis representing the window size. In this study, a dense sampling method is proposed that extends the sampling space from one to two dimensions. The new space consists of time and window axes. Dense sampling provides several desirable effects, such as a larger training dataset, an intra-model ensemble, model-agnosticism, and an easier setting of the optimal window. The experiments were conducted using four real datasets: Bitcoin price, influenza-like illness, household electric power consumption, and wind speed. The mean absolute percentage error was measured extensively in terms of varying window sizes, horizons, and lengths of time series. The resulting data showed that dense sampling significantly and consistently outperformed sparse sampling. The source codes and datasets are available at https://github.com/isoh24/Dense-sampling-timeseries.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.