A Composite Quantile Fourier Neural Network for Multi-Step Probabilistic Forecasting of Nonstationary Univariate Time Series
This work addresses the challenging problem of nonparametric probabilistic forecasting for time series, which is incremental as it applies an extrapolation-based approach not previously used in this context.
The authors tackled probabilistic forecasting of nonstationary univariate time series by proposing a composite quantile Fourier neural network, which achieved high-quality and accurate predictions validated on eight real-world datasets against nine benchmark methods.
Point forecasting of univariate time series is a challenging problem with extensive work having been conducted. However, nonparametric probabilistic forecasting of time series, such as in the form of quantiles or prediction intervals is an even more challenging problem. In an effort to expand the possible forecasting paradigms we devise and explore an extrapolation-based approach that has not been applied before for probabilistic forecasting. We present a novel quantile Fourier neural network is for nonparametric probabilistic forecasting of univariate time series. Multi-step predictions are provided in the form of composite quantiles using time as the only input to the model. This effectively is a form of extrapolation based nonlinear quantile regression applied for forecasting. Experiments are conducted on eight real world datasets that demonstrate a variety of periodic and aperiodic patterns. Nine naive and advanced methods are used as benchmarks including quantile regression neural network, support vector quantile regression, SARIMA, and exponential smoothing. The obtained empirical results validate the effectiveness of the proposed method in providing high quality and accurate probabilistic predictions.