Towards a Unified Multidimensional Explainability Metric: Evaluating Trustworthiness in AI Models

Title: Towards a Unified Multidimensional Explainability Metric: Evaluating Trustworthiness in AI Models Source: arxiv A multidimensional framework evaluates XAI methods like LIME and SHAP by scoring fidelity, simplicity, and stability across different models and datasets. This approach utilizes an offline knowledge base of model metadata to estimate explainability scores for unseen datasets and architectures.