Causal Triple Attention Time Series Forecasting

29 Sep 2021  ·  Zhixuan Chu, Tan Yan, Yue Wu, Yi Xu, Cheng Zhang, Yulin kang ·

Time series forecasting has historically been a key area of academic research and industrial applications. In multi-horizon and multi-series forecasting tasks, accurately capturing the local information in a sequence and effectively sharing global information across different sequences are very challenging, due to the complex dependencies over time in a long sequence and the heterogeneous nature across multiple time series. In this paper, from the perspective of causal inference, we give a theoretical analysis of these difficulties and establish a causal graph to identify the confounding relationship that generates harmful bias and misleads the time series model to capture the spurious correlations. We propose a causal triple attention time series forecasting model with three interpretable attention modules, which leverages the front-door adjustment to remove the confounding effect and help the model effectively utilize the local and global temporal information. We evaluate the performance of our model on four benchmark datasets and the results demonstrate the superiority over the state-of-the-art methods.

PDF Abstract

Datasets


  Add Datasets introduced or used in this paper

Results from the Paper


  Submit results from this paper to get state-of-the-art GitHub badges and help the community compare results to other papers.

Methods


No methods listed for this paper. Add relevant methods here