Fig. 12
From: Machine learning-based calibration of the GOCE satellite platform magnetometers
The figure shows the top 40 features used by the ML algorithm, sorted by their average impact on the final prediction for the calibrated value, as was calculated by the SHAP method. This was evaluated for the X, Y, and Z components of the measurements in the satellite frame where each feature’s contribution consists of the contribution to each component, distinguished by color. Note the different scale of the features on the left and right