Bridging classical, fuzzy, and approaches for missing data imputation: A unified comparative analysis

¹ Laboratory of Innovative Technologies and Computer Science, EST Fes, Sidi Mohamed Ben Abdellah University, Fes, Morocco.
² Laboratory of Engineering Sciences, FP Taza, Sidi Mohamed Ben Abdellah University, Fes, Morocco

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Published online: 24 June 2026

Abstract

Missing data remains a persistent challenge in data analysis, compromising statistical validity, predictive performance, and inferential robustness across diverse application domains. This paper presents a comprehensive comparative analysis of imputation methodologies, from foundational statistical approaches to state-of-the-art hybrid deep learning architectures. We establish a unified taxonomical organization that includes principal methodological families: classical statistical methods, machine learning and ensemble approaches, deep generative models, fuzzy logic systems, and hybrid architectures. Through rigorous empirical evaluation on benchmark datasets under varying missingness mechanisms—validated through convergent non-parametric statistical testing—we establish robust performance hierarchies and operational guidelines. Our findings demonstrate the statistical superiority of hybrid methodologies over both monolithic deep learning architectures and classical baselines, with hybrid approaches exhibiting enhanced precision and notable resilience to non-ignorable missingness. We further identify a structured Pareto frontier that balances computational efficiency and imputation accuracy, enabling context-specific method selection. Practical recommendations stratify methodological choices by operational context— critical applications that require maximal precision, real-time systems that demand computational efficiency, and scenarios characterized by complex missing mechanisms. This work provides theoretical foundations, empirical validation, and actionable decision frameworks for the rigorous handling of missing data in contemporary data science pipelines.