FDN: Interpretable Spatiotemporal Forecasting with Future Decomposition Networks
About
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.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Spatiotemporal forecasting | Wabash River (test) | MAE3.127 | 36 | |
| Spatiotemporal forecasting | E-PEMS-BAY (test) | MAE0.937 | 36 | |
| Forecasting | Solar-Energy (test) | MAE0.238 | 27 | |
| Spatiotemporal forecasting | Solar-Energy (test) | MAE0.238 | 27 | |
| Forecasting | E-PEMS-BAY Horizon 6 | MAE1.475 | 12 | |
| Forecasting | E-PEMS-BAY Horizon 12 | MAE1.792 | 12 | |
| Time Series Forecasting | Wabash River Basin Horizon 4 | MAE6.624 | 12 | |
| Time Series Forecasting | Wabash River Basin Horizon 7 | MAE9.122 | 12 | |
| Forecasting | E-PEMS-BAY Horizon 1 | MAE0.937 | 12 | |
| Time Series Forecasting | Wabash River Basin Horizon 1 | MAE3.127 | 12 |