<ul><li>Existing multivariate time series forecasting methods feed all variable histories into a unified model, hindering identification of variable-specific causal influences.</li><li>A new forecasting paradigm, CAIFormer, predicts each target variable separately using a Structural Causal Model and four sub-segments: endogenous, direct causal, collider causal, and spurious correlation.</li><li>CAIFormer consists of three components: Endogenous Sub-segment Prediction Block, Direct Causal Sub-segment Prediction Block, and Collider Causal Sub-segment Prediction Block.</li><li>Experiments on benchmark datasets show the effectiveness of CAIFormer in addressing the limitations of existing multivariate time series forecasting methods.</li></ul>

CAIFormer: A Causal Informed Transformer for Multivariate Time Series Forecasting

Discover more