A Systematic Review of Explainable Machine Learning Models for Distributed Big Data Systems
Big Data and Cognitive Computing · 14.11.2026 · 10 (11) · 389
DOI: 10.3390/bdcc10110389 ↗Kurz gefasst
Systematische Übersicht in der Zeitschrift Big Data and Cognitive Computing, online erschienen am 14.11.2026. Verfasst von Sadaf Zehra und 3 weiteren Autor:innen. Die Arbeit nennt 44 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NED University of Engineering and Technology.
Abstract (Originalsprache)
Background
With the continued integration of machine learning (ML) systems into distributed big-data environments, there is now an urgent need for explainability in order to achieve transparency, accountability, and trustworthiness in large intelligent systems. Cloud computing, edge intelligence, federated learning, blockchain-enabled infrastructures and Internet of Things (IoT) ecosystems are increasingly adding complexity to ML pipelines at scale, especially within high-stakes fields like cybersecurity, healthcare, finance and smart energy systems. In order to address the limitations of “Black Box” machine learning predictive models, Explainable Artificial Intelligence (XAI) has been developed to provide interpretable and human-understandable explanations of the complex decision-making processes inherent to distributed computing environments.
Objective
This research looks at how explainable ML models are used in current distributed Big Data Systems, with a focus on ML architecture and XAI technique(s) chosen; the types of distributed computing platforms available; evaluation methods; application areas; and open research questions related to these technologies.
Methods
A systematic literature review (large-scale) was performed according to the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) methodology. Peer-reviewed articles from 2020 to 2025 were obtained from major scientific databases and screened through multiple stages to include/exclude documents. Peer-reviewed Q1/Q2 journal articles applying explanation mechanisms in distributed or scalable ML environments formed the primary evidence base; seven studies, including two peer-reviewed main-conference papers and one survey, were retained under documented exception rules, and 13 further studies are reported as documented deviations. Several quality assessment measures, including six evaluation criteria (QA1–QA6), were utilized to evaluate the rigor of the study methods, whether or not the method was implemented as distributed technologies, whether or not explanations were validated, whether or not studies could be reproduced, and whether or not methodologies could be scaled to sufficient amounts of data to be practical for use. A narrative, frequency-based synthesis was performed, and a taxonomy of the included studies was developed.
Results
Thirty-two studies were reviewed, showing that research activity has increased markedly since 2020, with 26 of the 32 studies (81%) published in 2024–2025. Deep learning architectures (including CNNs, RNNs, and LSTMs) were the most frequently used model category (N = 27), followed by supervised machine learning (N = 17), ensemble/hybrid methods (N = 15), and Transformer-based models (N = 12). SHAP was the most frequently used XAI technique, appearing in 19 of the 32 studies (59%), followed by feature-importance methods (N = 14) and LIME (N = 10), while Integrated Gradients (N = 4) and Layer-wise Relevance Propagation (N = 3) were rarely used. Apache Spark, Hadoop, Kubernetes, and federated learning frameworks provided the basis for scalable distributed analytics; the most common computational environments used were cloud-based or edge (fog) configured computing platforms. The most frequent application domains were industrial/manufacturing, cybersecurity, and energy systems (N = 11 each), followed by healthcare (N = 10). While there were widespread reports of high levels of predictive performance for most of the studies, there were also a number of studies reporting significant drawbacks to explanation stability, computational efficiency, or level of feasibility for on-time deployment of solutions into production environments. Only 6 of the 32 studies reported explanation-specific measures such as fidelity, stability, or coherence.
Conclusions
This research shows that explainability in ML has emerged as an important aspect of distributed BDS, especially in environments where a high level of transparency, trust, and regulatory compliance is required. Although the explainability of ML has improved greatly over the years, many challenges still exist, such as the balance between accuracy and interpretability, the inability to scale post hoc explanatory methods, privacy issues in federated settings, and the lack of standardized evaluation methods for XAI. Furthermore, explainable ML at the resource-constrained edge and in TinyML devices continues to be an area of little exploration. Future work should focus on developing lightweight, intrinsically explainable ML models; federated explainable ML with privacy protection; hardware-accelerated XAI pipelines; mechanisms for causal reasoning; and standardizing XAI evaluation methods in order to build scalable, trustworthy, and human-centered intelligent systems for the next generation of distributed computing infrastructures.
Förderung
- NED University of Engineering and Technology
Zitieren
APA
Zehra, S., Farooq, M. U., Latif, M., & Khan, S. M. (2026). A Systematic Review of Explainable Machine Learning Models for Distributed Big Data Systems. Big Data and Cognitive Computing, 10(11), 389. https://doi.org/10.3390/bdcc10110389
Vancouver
Zehra S, Farooq MU, Latif M, Khan SM. A Systematic Review of Explainable Machine Learning Models for Distributed Big Data Systems. Big Data and Cognitive Computing. 2026;10(11):389. doi:10.3390/bdcc10110389
BibTeX
@article{Zehra2026,
author = {Zehra, Sadaf and Farooq, Muhammad Umer and Latif, Mustafa and Khan, Shariq Mahmood},
title = {A Systematic Review of Explainable Machine Learning Models for Distributed Big Data Systems},
journal = {Big Data and Cognitive Computing},
year = {2026},
volume = {10},
number = {11},
pages = {389},
doi = {10.3390/bdcc10110389}
}
EUVIMED