SHAP

A unified framework for feature-importance explanations of machine-learning models, introduced by Lundberg and Lee in 2017, grounded in Shapley values from cooperative game theory. For any model and input, SHAP assigns each feature a value representing its contribution to that particular prediction. SHAP is commonly used for tabular models (medical risk, eligibility determinations, moderation systems) that affect disabled users' lives, and for audio or language models where individual features — words, phonemes, or columns of a spectrogram — can be weighted. Accessibility limitations: SHAP outputs are typically visualised as bar charts or force plots that require visual perception, so presenting them accessibly requires translation to text or audio.