Tag
This paper introduces TSCOMP, a large-scale benchmark that systematically decomposes deep multivariate time-series forecasting methods into fine-grained components to enable automated model selection, outperforming complex holistic architectures.