This is a page that curates AI-related papers published worldwide. All content here is summarized using Google Gemini and operated on a non-profit basis. Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.
Investigating a Model-Agnostic and Imputation-Free Approach for Irregularly-Sampled Multivariate Time-Series Modeling
Created by
Haebom
Author
Abhilash Neog, Arka Daw, Sepideh Fatemi Khorasgani, Medha Sawhney, Aanish Pradhan, Mary E. Lofton, Bennett J. McAfee, Adrienne Breef-Pilz, Heather L. Wander, Dexter W Howard, Cayelan C. Carey, Paul Hanson, and Anuj Karpatne.
Outline
This paper addresses irregularly sampled multivariate time series (IMTS) modeling, which plays a crucial role in various applications where different variables may be missing at different points in time due to sensor failure or high data acquisition costs. Existing IMTS approaches either consider a two-stage imputation-modeling framework or employ architectures tailored to specific models and tasks. In this paper, we conduct a series of experiments to gain new insights into the performance of IMTS methods on classification and prediction tasks on a variety of semi-synthetic and real-world datasets. We also introduce Missing Feature-Aware Time Series Modeling (MissTSM), a novel model-agnostic and imputation-free approach for IMTS modeling. MissTSM demonstrates competitive performance compared to other IMTS approaches under typical conditions, particularly in real-world IMTS applications with a high number of missing values and a lack of simple periodic structure in the data.
Takeaways, Limitations
•
Takeaways: We experimentally verified that MissTSM outperforms existing IMTS modeling methods on real-world data with many missing values. It is particularly effective on data without simple periodic structures. Because it is a model-independent and non-substitution-intensive approach, it is highly applicable to a wide range of models and data.
•
Limitations: The experimental results presented in this paper may be limited to specific datasets and settings. It is difficult to generalize that MissTSM performs well on all types of IMTS data. Further validation is needed using a wider range of datasets and experimental settings. Furthermore, analysis of MissTSM's computational cost and scalability is lacking.
