TimeFilter: Patch-Specific Spatial-Temporal Graph Filtration for Time Series Forecasting

Yifan Hu, Guibin Zhang, Peiyuan Liu, Disen Lan, Naiqi Li, Dawei Cheng, Tao Dai, Shu-Tao Xia, Shirui Pan

Abstract: Time series forecasting methods generally fall into two main categories: Channel Independent (CI) and Channel Dependent (CD) strategies. While CI overlooks important covariate relationships, CD captures all dependencies without distinction, introducing noise and reducing generalization. Recent advances in Channel Clustering (CC) aim to refine dependency modeling by grouping channels with similar characteristics and applying tailored modeling techniques. However, coarse-grained clustering struggles to capture complex, time-varying interactions effectively. To address these challenges, we propose TimeFilter, a GNN-based framework for adaptive and fine-grained dependency modeling. After constructing the graph from the input sequence, TimeFilter refines the learned spatial-temporal dependencies by filtering out irrelevant correlations while preserving the most critical ones in a patch-specific manner. Extensive experiments on 13 real-world datasets from diverse application domains demonstrate the state-of-the-art performance of TimeFilter. The code is available at https://github.com/TROUBADOUR000/TimeFilter.

Lay Summary:

Forecasting future values in time series data is challenging because of the complex dependencies within the data. Traditional forecasting methods often struggle to balance capturing interdependencies among different data channels and avoiding irrelevant or noisy relationships.

We propose a new method called TimeFilter to address these challenges. It constructs a graph-based framework that adapts to the dependencies between different time periods and data channels. TimeFilter segments the input time series into patches and constructs a spatial-temporal graph, then filters out irrelevant correlations through patch-specific filtering, preserving only the most critical dependencies.

Experiments on multiple real-world datasets show that TimeFilter performs better than existing methods, making it a promising advancement for predicting future trends in various fields

Link To Code: https://github.com/TROUBADOUR000/TimeFilter

Primary Area: Deep Learning->Sequential Models, Time series

Keywords: Time Series Forecasting

Submission Number: 2565