Explainable AI-Based Data Analytics For Transparent and Reliable Predictive Decision-Making

Abstract

Artificial intelligence (AI) is increasingly being applied to predictive decision-making, although limited model interpretability may create challenges related to transparency, accountability, and understanding of automated outcomes. This study reviewed and synthesized reported evidence concerning explainable artificial intelligence (XAI)-based data analytics, with particular attention to predictive performance and the quality of generated explanations. Four machine-learning algorithms were considered using reported five-fold cross-validation results, whereas an explainable AI framework was examined through established classification performance measures. Shapley Additive Explanations (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), and Improved Local Interpretable Model-Agnostic Explanations (ILIME) were further compared according to local and global fidelity, stability, and monotonicity measures. Among the reviewed machine-learning approaches, XGBoost demonstrated the strongest overall predictive performance, while the evaluated explainable framework achieved high performance across the reported classification measures. SHAP demonstrated comparatively stronger results across the assessed explanation-quality indicators than LIME and ILIME. These findings suggest that predictive capability and interpretability can be addressed together within AI-based decision-support systems. Integrating explanation mechanisms with predictive models may therefore contribute to more transparent and reliable analytical processes, particularly in settings where understanding model outputs is important for informed decision-making. The synthesis also highlights the value of evaluating both predictive effectiveness and explanation quality when considering XAI approaches for practical AI applications.