Additivity says that attribution for a combined value function equals the sum of the separate attributions. If two models or value functions are combined, each feature's overall Shapley attribution can be obtained by adding its attributions from the individual parts. section 9.4.2 section 9.6
From separate models to an ensemble
Suppose an ensemble contains several base estimators. Compute the Shapley values for a feature in each estimator, then add those values to obtain the feature's attribution for the ensemble. This makes additive explanations practical for ensemble methods without treating the ensemble as an unrelated black box. section 9.4.2 section 9.6
Figure 9.20 A force plot to visualize Shapley values. The features pointing to the right push the prediction on this example (0.52) to be higher than the average prediction (-0.194). The features pointing to the left push the prediction lower and closer to base value. The contributions add up to the overall prediction, which is higher than the base value for this example. The feature values of the example being explained are shown along with the features under the force plot.
A force plot makes additivity visible: feature contributions push a prediction above or below a base value, and the contributions sum to the overall prediction. section 9.4.2
