Explainable Artificial Intelligence and Hybrid Machine Learning Models for Transparent Intelligent Decision Support: A Structured Review of Methods, Architectures, and Deployment Challenges

Ananya R. Verma, Rohit K. Sahu, Priya M. Tiwari, Deepak S. Yadav

Abstract


ABSTRACT
Machine learning models now inform decisions in clinical triage, credit underwriting, and industrial maintenance, yet the most accurate models are often opaque, which limits accountability, regulatory compliance, and user trust. This review examines how Explainable Artificial Intelligence (XAI) and hybrid machine learning architectures can be combined to provide transparent intelligent decision support. A structured narrative review of representative literature from 2015 to 2025 was conducted and coded on four dimensions: explanation mechanism, model architecture, evaluation level, and application domain. The synthesis shows that post-hoc attribution methods such as LIME and SHAP dominate practice [1], [2], but their fidelity and stability are conditional and can be undermined [3], [4]. Hybrid designs that pair an interpretable glass-box model with a high-capacity learner, or that route predictions through human-understandable concepts, narrow the accuracy–transparency gap while keeping explanations faithful [5], [6]. The paper proposes a layered reference framework, identifies five recurring hybrid design patterns, maps them to clinical, financial, and industrial contexts, and outlines open problems in evaluation, causal grounding, and governance. The evidence supports hybrid-by-design transparency for high-stakes decision support.

KEYWORDS: Explainable artificial intelligence, Hybrid machine learning, Decision support systems, Interpretability, SHAP, LIME, Glass-box models, Neurosymbolic AI, Trustworthy AI, Counterfactual explanations

Full Text:

PDF 204-216