Figure 14From: Machine learning applied to simulations of collisions between rotating, differentiated planetsSHAP values for a selected subset of post-impact parameteres. The SHAP values are a useful metric for explaining how data-driven models classify or predict collision outcomes. On the x-axis, the SHAP value quantifies the magnitude of the contribution by each pre-impact quantity. Negative SHAP values push the value of the post-impact parameter lower, whereas positive SHAP values push the value higher. The normalized value of the pre-impact parameter (ordered along the y-axis) is indicated by color, with bluer values indicating lower pre-impact parameter values and higher values redBack to article page