Combining Multiple Imputation and Deep Learning to Handle Missing Values
Journal of Economics, Business, and Commerce · 06.06.2026 · 3 (1) · 226-235
DOI: 10.69739/jebc.v3i1.1980 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Economics, Business, and Commerce, online erschienen am 06.06.2026. Verfasst von Hadiya H. Matrood. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
This article reviews a recent study on the missing data problem and its solution using Multiple Imputation (MI) approaches with the help of deep learning models. This study demonstrates the use of modern neural network-based approaches (such as generative adversarial networks (GANs) and autoencoders) to more accurately estimate missing values than conventional approaches, especially when patterns in large, complex, and non-linear datasets are considered. This study also discusses the types of missing data (MCAR, MAR, and NMAR) and their effects on statistical analyses. The research found that combining deep learning with Multiple Imputation not only improves data imputation but also more accurately estimates uncertainty, resulting in more accurate final outcomes. Furthermore, the study emphasizes the critical importance of selecting appropriate imputation strategies tailored to specific dataset characteristics and missingness mechanisms, as poor strategy selection can introduce systematic bias and substantially compromise the validity of downstream statistical inferences. The authors also underscore that the choice of imputation model must be guided not only by the volume and dimensionality of the data but also by the underlying distributional assumptions, since misalignment between the imputation model and the true data-generating process can lead to inflated type I error rates and reduced statistical power. Additionally, the integration of deep learning architectures into the MI framework offers promising avenues for handling heterogeneous data types, including mixed categorical and continuous variables, which are commonly encountered in clinical, social, and behavioral research settings. Finally, this study highlighted some of the existing problems, including computational cost and interpretability, and future research opportunities, such as the use of reinforcement learning and graphs in the embedding processes.
Zitieren
APA
Matrood, H. H. (2026). Combining Multiple Imputation and Deep Learning to Handle Missing Values. Journal of Economics, Business, and Commerce, 3(1), 226-235. https://doi.org/10.69739/jebc.v3i1.1980
Vancouver
Matrood HH. Combining Multiple Imputation and Deep Learning to Handle Missing Values. Journal of Economics, Business, and Commerce. 2026;3(1):226-235. doi:10.69739/jebc.v3i1.1980
BibTeX
@article{Matrood2026,
author = {Matrood, Hadiya H.},
title = {Combining Multiple Imputation and Deep Learning to Handle Missing Values},
journal = {Journal of Economics, Business, and Commerce},
year = {2026},
volume = {3},
number = {1},
pages = {226--235},
doi = {10.69739/jebc.v3i1.1980}
}
EUVIMED