IBN: An Interpretable Bidirectional-Modeling Network for Multivariate Time Series Forecasting with Variable Missing

📅 2025-09-09
📈 Citations: 0
Influential: 0
📄 PDF
🤖 AI Summary
Multivariate time series forecasting faces challenges in modeling spatial correlations, capturing temporal patterns, and ensuring interpretability under missing variable observations. To address these issues, we propose the Interpretable Bidirectional Network (IBN), which integrates Uncertainty-Aware Imputation (UAI), Gaussian Kernel Graph Convolution (GGCN), and a bidirectional recurrent architecture. UAI leverages Monte Carlo Dropout to quantify reconstruction uncertainty and guide imputation via uncertainty-weighted aggregation; GGCN adaptively learns dynamic inter-variable dependencies through kernelized graph convolutions; and the bidirectional structure explicitly models both forward and backward temporal dependencies. This design jointly enhances prediction accuracy, robustness, and interpretability. Extensive experiments demonstrate that IBN consistently outperforms state-of-the-art baselines—including GinAR—across diverse missingness rates, particularly under high missingness scenarios, while significantly improving prediction stability and reliability.

Technology Category

Application Category

📝 Abstract
Multivariate time series forecasting (MTSF) often faces challenges from missing variables, which hinder conventional spatial-temporal graph neural networks in modeling inter-variable correlations. While GinAR addresses variable missing using attention-based imputation and adaptive graph learning for the first time, it lacks interpretability and fails to capture more latent temporal patterns due to its simple recursive units (RUs). To overcome these limitations, we propose the Interpretable Bidirectional-modeling Network (IBN), integrating Uncertainty-Aware Interpolation (UAI) and Gaussian kernel-based Graph Convolution (GGCN). IBN estimates the uncertainty of reconstructed values using MC Dropout and applies an uncertainty-weighted strategy to mitigate high-risk reconstructions. GGCN explicitly models spatial correlations among variables, while a bidirectional RU enhances temporal dependency modeling. Extensive experiments show that IBN achieves state-of-the-art forecasting performance under various missing-rate scenarios, providing a more reliable and interpretable framework for MTSF with missing variables. Code is available at: https://github.com/zhangth1211/NICLab-IBN.
Problem

Research questions and friction points this paper is trying to address.

Addresses missing variables in multivariate time series forecasting
Overcomes interpretability limitations in existing attention-based methods
Enhances spatial-temporal modeling with bidirectional uncertainty-aware reconstruction
Innovation

Methods, ideas, or system contributions that make the work stand out.

Uncertainty-Aware Interpolation with MC Dropout
Gaussian kernel-based Graph Convolution modeling
Bidirectional recurrent unit enhancing temporal dependencies
🔎 Similar Papers
No similar papers found.
S
Shusen Ma
University of Science and Technology of China, Hefei, China
T
Tianhao Zhang
University of Science and Technology of China, Hefei, China
Q
Qijiu Xia
University of Science and Technology of China, Hefei, China
Yun-Bo Zhao
Yun-Bo Zhao
University of Science and Technology of China
Human-Machine SystemsSmart ManufacturingNetworked Control Systems