Shapley Values quantify what in a model?

Shapley Values quantify what in a model?

• A. The speed of model inference
• B. The contribution of each feature to the prediction
• C. The overall accuracy of the model
• D. The number of features used

Answer: (B)

Shapley values attribute a model’s prediction to its input features. They measure how much each feature contributes to pushing the prediction away from a baseline (often the average prediction) by averaging the feature’s marginal impact across all possible subsets of other features. This creates a fair breakdown where the contributions from all features add up to the difference between the actual prediction and the baseline. It also preserves useful properties: a feature that never changes the output gets a zero value; features with identical effects receive equal shares; and the total of all feature contributions matches the prediction. Shapley values can explain a single prediction (local) or be aggregated for a sense of global behavior. They are not about how fast the model runs, the overall accuracy, or how many features are used.

Shapley values attribute a model’s prediction to its input features. They measure how much each feature contributes to pushing the prediction away from a baseline (often the average prediction) by averaging the feature’s marginal impact across all possible subsets of other features. This creates a fair breakdown where the contributions from all features add up to the difference between the actual prediction and the baseline. It also preserves useful properties: a feature that never changes the output gets a zero value; features with identical effects receive equal shares; and the total of all feature contributions matches the prediction. Shapley values can explain a single prediction (local) or be aggregated for a sense of global behavior. They are not about how fast the model runs, the overall accuracy, or how many features are used.