Fig. 2

Prediction performances of the model for predicting the presence or absence of clams in a voxel. The model was trained and evaluated using Dataset 1 (presence or absence of clams). (a) Predicted distribution examples for each data type: control data containing sand only (C), data containing asari clams and sand (A), data containing a mixture of mussels and sand (M), and data containing a mixture of clams, mussels, and other materials (AM). Note that the presence or absence of clams is predicted from the corresponding voxel data, not from these photo images. The prediction is performed for the voxel corresponding to each square on these images. Regions marked with × indicate failed predictions, and regions marked with ✔ indicate successful predictions. (b) and (c) The prediction accuracy and the ROC curve for each voxel. (d) The prediction accuracy for each bucket.