Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Fig 1.

Proposed Methodology.

More »

Fig 1 Expand

Fig 2.

Distribution of export values visualized with and without log scale.

Left: original scale; Right: log scale.

More »

Fig 2 Expand

Fig 3.

Exploring relationships between export value and key features.

The figure shows how export value relates to important variables in the dataset.

More »

Fig 3 Expand

Fig 4.

Average export value trends across feature levels, illustrating how export values vary with key features in the dataset.

More »

Fig 4 Expand

Fig 5.

Feature importance ranking using the Random Forest Regressor for predicting export values.

The plot shows which features contributed most to the model’s predictions.

More »

Fig 5 Expand

Fig 6.

Graph structure and connectivity analysis of the country trade network.

More »

Fig 6 Expand

Fig 7.

Proposed Graph Neural Network (GNN) architecture.

More »

Fig 7 Expand

Fig 8.

Learning curves of the proposed model: (a) training and validation loss (MSE), (b) validation RMSE, and (c) validation MAE over 250 epochs.

More »

Fig 8 Expand

Fig 9.

Training behavior of the proposed GNN model for trade balance forecasting: (left) validation R2 score across epochs, achieving a stable value of approximately 0.91, and (right) step-wise learning rate decay used during training.

More »

Fig 9 Expand

Table 1.

Performance comparison of the proposed model against baseline methods.

More »

Table 1 Expand

Fig 10.

Residual distribution analysis of the proposed GNN model for trade balance forecasting.

The top row shows predicted versus actual trade balance for the training, validation, and test sets. The middle row presents residual plots, while the bottom row illustrates the residual error distributions along with MAE values for each dataset.

More »

Fig 10 Expand

Table 2.

Paired t-test results comparing the proposed GNN model with baseline methods.

More »

Table 2 Expand

Table 3.

Wilcoxon signed-rank test results comparing the proposed GNN model with baseline methods.

More »

Table 3 Expand

Table 4.

Sensitivity analysis results of the proposed GNN model across baseline, geographical, and temporal settings.

More »

Table 4 Expand

Table 5.

Top 20 features ranked by mean absolute SHAP values.

More »

Table 5 Expand

Fig 11.

SHAP summary plot showing the global feature importance and contribution patterns for trade balance prediction using the proposed GNN model.

More »

Fig 11 Expand