EXPLAINABLE ARTIFICIAL INTELLIGENCE FOR RELIABLE DECISION SUPPORT

Article Information

Abstract

Machine learning predictions are increasingly accompanied by explanations intended to support human judgment. The usefulness of an explanation, however, depends on its faithfulness to the model, the assumptions used to generate it, and the task performed by its audience. This narrative review examines local surrogate explanations, additive feature attribution, counterfactual explanations, interpretable models, and selected methods for evaluating explanation reliability. Six foundational publications provide the basis for an analytical comparison. The paper develops a conceptual evaluation framework for educational decision support that separates predictive quality, explanation fidelity, stability, and human decision performance. It argues that an understandable explanation cannot establish the correctness, fairness, or causal validity of a prediction. Aproposed pilot compares an interpretable baseline with a more complex model and evaluates whether explanations help users detect errors and respond appropriately to uncertainty. No new dataset, experiment, or measured improvement is reported. The principal contribution is a practical account of what should be tested before explanations are used to support consequential institutional decisions.

Keywords

DOI Active Record

This scholarly article has been successfully registered and assigned a persistent DOI through DOIGLOBAL.

Research Tools

Share This Research

Recommended Citation