International Research Journal of Education and Technology

Modern artificial intelligence systems are increasingly deployed in consequential domains such as criminal justice, healthcare resource allocation, and financial lending. Despite impressive predictive accuracy, these systems frequently encode and amplify societal biases present in their training data, resulting in disparate outcomes across demographic groups. This paper introduces the Ethical Bias Heatmap Generator (EBHG), a multi-dimensional fairness auditing framework that produces layered visual representations of bias across model architectures, demographic intersections, and decision boundaries. EBHG integrates statistical parity analysis, equalized odds evaluation, and counterfactual fairness testing into a unified heatmap visualization that reveals bias concentrations invisible to conventional scalar fairness metrics. We validate EBHG across three domains—criminal recidivism prediction, medical triage, and automated hiring—demonstrating 94% bias detection coverage compared to 67% for single-metric approaches. User studies with 38 ML practitioners confirm that heatmap-based auditing reduces bias identification time by 42% and improves remediation accuracy by 31% over tabular reporting methods. These results establish EBHG as a practical instrument for responsible AI governance. 

Index Terms—Ethical AI, bias detection, fairness metrics, heatmap visualization, algorithmic auditing, responsible AI, intersectional bias, counterfactual fairness.