Missing Data Imputation Methods Systematically Reviewed
A thorough examination of methods for imputing missing data has been released on arXiv, consolidating years of disjointed research from various fields including healthcare, bioinformatics, social science, e-commerce, and industrial monitoring. This paper delves into essential topics such as the mechanisms of missingness, the distinction between single and multiple imputation, and various imputation objectives. It classifies techniques ranging from traditional methods like regression and the EM algorithm to contemporary strategies such as low-rank and high-rank matrix completion, as well as deep learning approaches (including autoencoders, GANs, and diffusion models). The review seeks to bridge statistical principles with the latest developments in machine learning, highlighting the urgent need for an integrated framework in data science.
Key facts
- The review covers missing data imputation methods across healthcare, bioinformatics, social science, e-commerce, and industrial monitoring.
- It synthesizes literature that has been fragmented across different fields.
- Core concepts include missingness mechanisms, single vs. multiple imputation, and imputation goals.
- Classical techniques covered: regression, EM algorithm.
- Modern approaches include low-rank and high-rank matrix completion, autoencoders, GANs, and diffusion models.
- The paper connects statistical foundations with machine learning advances.
- Published on arXiv with ID 2511.01196.
- The review is interdisciplinary and cross-task.
Entities
Institutions
- arXiv