In recent years, the exponential growth of high-dimensional datasets across fields such as genomics, finance, and cybersecurity has amplified the need for efficient and interpretable machine learning systems. While deep learning models demonstrate remarkable accuracy in pattern recognition tasks, they often lack transparency, posing challenges for trust, accountability, and regulatory compliance. Explainable Artificial Intelligence (XAI) has emerged as a critical research frontier aimed at bridging this interpretability gap. However, most standalone XAI models sacrifice performance for transparency, especially in high-dimensional spaces. This research investigates the efficiency of hybrid XAI models—those that integrate interpretable layers, post-hoc explanation methods, or modular learning structures—with conventional high- performance models to balance accuracy and interpretability. The study adopts a comparative experimental approach using datasets from image recognition and bioinformatics, applying hybrid models such as SHAP-integrated convolutional neural networks (CNNs) and attention-guided recurrent networks. Key performance indicators include classification accuracy, feature importance fidelity, and explanation stability. Statistical tools such as ANOVA and confidence interval analysis are employed to evaluate significance across models. Findings suggest that hybrid models can retain competitive accuracy while offering clearer feature-level insights, thereby enhancing stakeholder trust and model accountability. Furthermore, these models demonstrate potential in uncovering latent patterns often missed by conventional dimensionality reduction techniques. The study underscores the viability of hybrid XAI models in critical decision-making domains, advocating for their broader adoption in real-world high-dimensional data mining tasks (Doshi-Velez & Kim, 2017).
Ur-Rahaman et al. (Fri,) studied this question.