STOAT: Spatial-Temporal Probabilistic Causal Inference Network

The introduction of the STOAT (Spatial-Temporal Probabilistic Causal Inference Network) framework represents a pivotal development in forecasting spatial-temporal causal time series (STC-TS). Traditional methods often treat spatial and temporal dynamics separately, limiting their effectiveness. STOAT addresses this by integrating a spatial relation matrix to capture interregional dependencies, thereby improving causal effect estimation. Its performance was validated through experiments on COVID-19 data from six countries, where it demonstrated superior accuracy compared to state-of-the-art models like DeepAR and DeepVAR. This advancement not only enhances predictive capabilities but also offers a generalizable framework for complex spatial-temporal tasks, bridging the gap between causal inference and geospatial probabilistic forecasting.