Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BRATI: Bidirectional Recurrent Attention for Time-Series Imputation

Armando Collado-Villaverde, Pablo Muñoz, Maria D. R-Moreno

TL;DR

BRATI tackles missing data in multivariate time-series by integrating bidirectional recurrent processing with self-attention to model short- and long-range dependencies and cross-feature correlations. The model employs two imputation blocks running in opposite temporal directions, a learned weighted fusion, embedding with positional encoding, and a joint training regime with Masked Imputation, Observed Reconstruction, and Consistency objectives. Across PhysioNet, Water Quality, and Space Weather datasets, BRATI consistently outperforms SoA baselines, with pronounced gains on complex random-length missing patterns and MNAR-like scenarios; Space Weather provides a challenging MNAR benchmark. The work demonstrates practical impact for robust imputation in diverse domains and introduces Space Weather as a realistic dataset for MNAR imputation research.

Abstract

Missing data in time-series analysis poses significant challenges, affecting the reliability of downstream applications. Imputation, the process of estimating missing values, has emerged as a key solution. This paper introduces BRATI, a novel deep-learning model designed to address multivariate time-series imputation by combining Bidirectional Recurrent Networks and Attention mechanisms. BRATI processes temporal dependencies and feature correlations across long and short time horizons, utilizing two imputation blocks that operate in opposite temporal directions. Each block integrates recurrent layers and attention mechanisms to effectively resolve long-term dependencies. We evaluate BRATI on three real-world datasets under diverse missing-data scenarios: randomly missing values, fixed-length missing sequences, and variable-length missing sequences. Our findings demonstrate that BRATI consistently outperforms state-of-the-art models, delivering superior accuracy and robustness in imputing multivariate time-series data.

BRATI: Bidirectional Recurrent Attention for Time-Series Imputation

TL;DR

BRATI tackles missing data in multivariate time-series by integrating bidirectional recurrent processing with self-attention to model short- and long-range dependencies and cross-feature correlations. The model employs two imputation blocks running in opposite temporal directions, a learned weighted fusion, embedding with positional encoding, and a joint training regime with Masked Imputation, Observed Reconstruction, and Consistency objectives. Across PhysioNet, Water Quality, and Space Weather datasets, BRATI consistently outperforms SoA baselines, with pronounced gains on complex random-length missing patterns and MNAR-like scenarios; Space Weather provides a challenging MNAR benchmark. The work demonstrates practical impact for robust imputation in diverse domains and introduces Space Weather as a realistic dataset for MNAR imputation research.

Abstract

Missing data in time-series analysis poses significant challenges, affecting the reliability of downstream applications. Imputation, the process of estimating missing values, has emerged as a key solution. This paper introduces BRATI, a novel deep-learning model designed to address multivariate time-series imputation by combining Bidirectional Recurrent Networks and Attention mechanisms. BRATI processes temporal dependencies and feature correlations across long and short time horizons, utilizing two imputation blocks that operate in opposite temporal directions. Each block integrates recurrent layers and attention mechanisms to effectively resolve long-term dependencies. We evaluate BRATI on three real-world datasets under diverse missing-data scenarios: randomly missing values, fixed-length missing sequences, and variable-length missing sequences. Our findings demonstrate that BRATI consistently outperforms state-of-the-art models, delivering superior accuracy and robustness in imputing multivariate time-series data.
Paper Structure (30 sections, 14 equations, 3 figures, 13 tables)

This paper contains 30 sections, 14 equations, 3 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Multivariate Time-Series with Missing values
  5. Joint-optimization Training approach
  6. BRATI
  7. Embedding layer and Positional Encoder
  8. Attention
  9. GRU
  10. Feed-Forward network
  11. Imputation blocks
  12. Weighted Combination
  13. Loss functions
  14. Experiments
  15. Artificial generation of missing values
  16. ...and 15 more sections

Figures (3)

  • Figure 1: Example of the missing plasma (red line) values (shaded in gray) during a geomagnetic storm, when the intensity of the storm reaches its peak (represented by the geomagnetic index SYM-H, in black).
  • Figure 2: Graphical summary of the joint Training approach using the Masked Imputation Loss ($L_{\text{MIL}}$) and the Observed Reconstruction Loss ($L_{\text{ORL}}$).
  • Figure 3: BRATI Imputation architecture. The dotted purple rectangle represents the forward imputation block, the dashed yellow rectangle represents the backwards imputation blocks. The blue rounded rectangles represent operation with weights that can be optimized whereas the green straight rectangles represent operation without optimizable parameters.