Fig. 1. (a, b) Curves showing the change in the mean absolute error (MAE) deviation of the total training data and validation set as the proportion of the training set varies, (c, d) Learning curves (loss vs. number of decision trees) for the training set and validation set when the training-to-validation ratio is 4:1.

Fig. 2. Density distribution of MAE for different modes Results from 500 runs for each mode (Mode_gp1, Mode_W and Mode_BF) and from ECC model are displayed. Dashed lines denote a Gaussian fit to the distribution. In each run, the 2 610 fusion reactions were randomly split into training and validation sets at a ratio of 4:1.

Fig. 3. The fusion cross sections predicted by different modes and by Wong formula (color online)Black dots represent experimental data taken from Refs.[47-53].

Fig. 4. The barrier distribution predicted by different models with pentagrams representing experimental data

Fig. 5. The ranking of important features was obtained using SHAP. Each row represents a feature, and the x-axis indicates the SHAP value, showing the importance of the feature for specific predictions. The pie chart illustrates the contribution proportions of different feature categories. The figure displays three machine learning modes, with training data derived from the training set.