## Heatmap: Privacy Budget Impact on Model Performance ### Overview The heatmap illustrates the impact of varying privacy budgets on the performance of different machine learning models. The models compared are SDGNN, SIGAT, SGCN, GAP, LSN, and ASGL (Proposed). The performance is measured using AUC (Area Under the Curve) on a binary classification task. ### Components/Axes - **Rows**: Represent different privacy budgets (c = 1, 2, 3, 4, 5, 6). - **Columns**: Represent different models (SDGNN, SIGAT, SGCN, GAP, LSN, ASGL). - **Color Scale**: A gradient from green (lower AUC) to red (higher AUC) indicates the performance of each model at each privacy budget. ### Detailed Analysis or ### Content Details - **SDGNN**: Shows a consistent increase in AUC as the privacy budget increases, indicating improved performance. - **SIGAT**: Also shows an increase in AUC with higher privacy budgets, but the improvement is less pronounced compared to SDGNN. - **SGCN**: Displays a moderate increase in AUC with privacy budgets, but the performance is generally lower than SDGNN and SIGAT. - **GAP**: Shows a slight increase in AUC with privacy budgets, but the performance is not significantly better than the other models. - **LSN**: Demonstrates a slight increase in AUC with privacy budgets, but the performance is not as high as SDGNN or SIGAT. - **ASGL (Proposed)**: Shows the highest AUC across all privacy budgets, indicating the best performance among the models tested. ### Key Observations - **Performance Improvement**: All models show an improvement in AUC as the privacy budget increases. - **ASGL's Superiority**: ASGL consistently outperforms the other models across all privacy budgets. - **Model Variability**: There is significant variability in performance among the models, with ASGL showing the most consistent improvement. ### Interpretation The heatmap suggests that increasing the privacy budget generally improves the performance of the models. ASGL, in particular, demonstrates the most significant improvement in performance as the privacy budget increases. This could be due to ASGL's ability to balance privacy and performance more effectively. The other models, while also improving, do not match ASGL's performance at the same privacy budget levels. This finding is crucial for applications where privacy is a concern, as it indicates that ASGL might be a better choice for such scenarios.