The exponential growth of high-dimensional data across domains such as healthcare, finance, and industrial automation necessitates robust models capable of both accurate predictions and explainable decision-making. Traditional machine learning models, including Support Vector Machines (SVM), Random Forests, and ensemble techniques, often struggle to generalize effectively when feature spaces become large and noisy. Deep learning models such as CNNs, LSTMs, and Transformers, while powerful, typically operate as black boxes, providing limited interpretability and weak uncertainty quantification. To address these challenges, this study introduces the MetaXAI Adaptive Learning Model, a meta-learning-based explainable artificial intelligence framework designed for complex algorithmic decision-making in high-dimensional environments. MetaXAI integrates autoencoder-based dimensionality reduction, attention-driven embedding, and a Model-Agnostic Meta-Learning (MAML) engine with Gradient Episodic Memory (GEM) for task adaptation. Furthermore, it combines SHAP, LIME, and counterfactual explanation modules to enhance interpretability. Results demonstrate that MetaXAI achieves an accuracy of 98.91%, outperforming traditional models such as SVM (91.3%), XGBoost (93.8%), and Transformer (94.8%). It also exhibits superior explainability (fidelity 98.2%) and uncertainty calibration (Expected Calibration Error 0.9%). Federated deployment further ensures privacy preservation with minimal communication overhead. The findings suggest that MetaXAI bridges the gap between predictive performance and interpretability, making it suitable for critical domains where both accuracy and transparency are paramount.
Devi et al. (Fri,) studied this question.