Abstract—Missing data poses a significant challenge in multivariate time series, disrupting continuity and leading to biases in analysis. Addressing these gaps is essential, as incomplete data can undermine the reliability of models in various applications. To overcome this challenge, we propose Fusion-Based Multivariate Time Series Imputation (Fusion-MTSI), a novel imputation method that addresses the limitations of traditional approaches, which often struggle to capture complex temporal and cross-feature relationships. Fusion-MTSI overcomes this limitation by leveraging both feature-wise and point-wise comparisons, enabling the detection of broad patterns and subtle temporal variations across features. By relying only on the intrinsic characteristics of data, Fusion-MTSI achieves effective imputation without requiring domain-specific knowledge. Experimental results across six real-world datasets demonstrate that Fusion-MTSI outperforms conventional methods, achieving up to a 31% reduction in Mean Squared Error, making it a robust, adaptable choice for diverse applications.
 
Keywords—missing data imputation, multivariate time series, feature relation, temporal causality, machine learning


Copyright © 2025 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).