FDN: Interpretable Spatiotemporal Forecasting with Future Decomposition Networks
It addresses the need for interpretable forecasts in spatiotemporal systems, offering a practical alternative to black-box models.
The paper introduces the Future Decomposition Network (FDN) for interpretable spatiotemporal forecasting, achieving competitive accuracy with state-of-the-art methods while using significantly less memory and runtime across hydrologic, traffic, and energy datasets.
Spatiotemporal systems comprise a collection of spatially distributed yet interdependent entities each generating unique dynamic signals. Highly sophisticated methods have been proposed in recent years delivering state-of-the-art (SOTA) forecasts but few have focused on interpretability. To address this, we propose the Future Decomposition Network (FDN), a novel forecast model capable of (a) providing interpretable predictions through classification (b) revealing latent activity patterns in the target time-series and (c) delivering forecasts competitive with SOTA methods at a fraction of their memory and runtime cost. We conduct comprehensive analyses on FDN for multiple datasets from hydrologic, traffic, and energy systems, demonstrating its improved accuracy and interpretability.