Analysis using deep neural networks and statistical predictors improves forecast accuracy in various datasets, indicating potential for applications in data-scarce scenarios.
Key Points
The hybrid model improved multi-horizon forecasting, enhancing predictive accuracy in volatile and intermittent datasets.
Using statistical forecasters as covariates in deep neural networks reduced SMAPE by approximately 33% on synthetic and stock market datasets.
Observational analysis covered four datasets including M5, Stallion, Stock Market, and Synthetic, highlighting diverse performance improvements.
These findings support the integration of statistical predictions to boost accuracy in forecasting, particularly for challenging time series data.