Figure 1.
Figure 2.
Figure 3.
Figure 4.
Figure 5.
Figure 6.
Figure 7.
Figure 8.
Figure 9.
Figure 10.
Figure 11.
Figure 12.
Figure 13.
Figure 14.
Figure 15.
Figure 16.
Figure 17.
Figure 18.
Figure 19.
Figure 20.
Figure 21.
Prediction results on the device selection task from [13]_
| Model | Accuracy (%) | Top-3 (%) | F1 (%) |
|---|---|---|---|
| DRAN-Net | 98.82 ± 0.41 | 100 ± 0.00 | 98.71 ± 0.50 |
| Random Forest | 98.15 ± 0.63 | 100 ± 0.00 | 97.01 ± 1.00 |
Prediction results after removal of four lower-importance features_
| Model+Representation Set | Accuracy(%) | Top-3(%) | F1(%) |
|---|---|---|---|
| Random Forest+T-Pre+AC | 78.13 ± 0.08 | 95.84 ± 0.11 | 80.93 ± 0.22 |
| Random Forest+N-Pre | 78.88 ± 0.07 | 95.87 ± 0.16 | 81.90 ± 0.14 |
| Gradient Boosting+T-Pre+AC | 76.84 ± 0.19 | 94.91 ± 0.17 | 79.82 ± 0.20 |
| Gradient Boosting+N-Pre | 77.31 ± 0.05 | 95.24 ± 0.61 | 79.95 ± 0.27 |
| Decision Tree+T-Pre+AC | 73.05 ± 0.12 | 94.64 ± 0.22 | 76.14 ± 1.46 |
| Decision Tree+N-Pre | 73.47 ± 0.34 | 95.17 ± 0.64 | 74.35 ± 1.15 |
| Nearest Neighbor+T-Pre+AC | 72.48 ± 0.00 | 93.68 ± 0.00 | 77.13 ± 0.00 |
| Nearest Neighbor+N-Pre | 73.73 ± 0.00 | 93.34 ± 0.00 | 78.56 ± 0.00 |
Prediction results between the new model and the traditional representation combination_
| Model+Representation Set | Accuracy (%) | Top-3 (%) | F1 (%) |
|---|---|---|---|
| DRAN-Net+T-Pre | 82.32 ± 1.16 | 97.87 ± 0.48 | 82.18 ± 1.08 |
| Random Forest+T-Pre | 77.39 ± 0.05 | 95.54 ± 0.12 | 80.75 ± 0.12 |
| Gradient Boosting+T-Pre | 76.63 ± 0.21 | 94.58 ± 0.17 | 80.47 ± 0.76 |
| Decision Tree+T-Pre | 72.93 ± 0.10 | 95.17 ± 0.00 | 78.52 ± 0.20 |
| Nearest Neighbor+T-Pre | 72.48 ± 0.00 | 93.68 ± 0.00 | 77.13 ± 0.00 |
| Multilayer Perceptron+T-Pre | 65.35 ± 0.54 | 82.06 ± 1.18 | 58.49 ± 1.57 |
| Support Vector Machine+T-Pre | 62.49 ± 0.00 | 78.70 ± 0.00 | 69.51 ± 0.00 |
| Naive Bayes+T-Pre | 33.60 ± 0.00 | 54.08 ± 0.00 | 29.48 ± 0.00 |
Efficiency and effectiveness of different methods_
| Method | Average Time(s) | Time Ranges(s) | Accuracy (%) | Top-3 (%) |
|---|---|---|---|---|
| Exhaustive Search | 118 | 5-243 | 100 ± 0.00 | 100 ± 0.00 |
| DRAN-Net Prediction | 0.27 | 0.19-0.51 | 90.40± 0.80 | 99.40±0.49 |
Prediction results between the new representation and the traditional model combination_
| Model+Representation Set | Accuracy(%) | Top-3(%) | F1(%) |
|---|---|---|---|
| Random Forest+T-Pre | 77.39 ± 0.05 | 95.54 ± 0.12 | 80.75 ± 0.12 |
| Random Forest+N-Pre | 78.88 ± 0.07 | 95.87 ± 0.16 | 81.90 ± 0.14 |
| Gradient Boosting+T-Pre | 76.63 ± 0.21 | 94.58 ± 0.17 | 80.47 ± 0.76 |
| Gradient Boosting+N-Pre | 77.31 ± 0.05 | 95.24 ± 0.61 | 79.95 ± 0.27 |
| Decision Tree+T-Pre | 72.93 ± 0.10 | 95.17 ± 0.00 | 78.52 ± 0.20 |
| Decision Tree+N-Pre | 73.47 ± 0.34 | 95.17 ± 0.64 | 74.35 ± 1.15 |
| Nearest Neighbor+T-Pre | 72.48 ± 0.00 | 93.68 ± 0.00 | 77.13 ± 0.00 |
| Nearest Neighbor+N-Pre | 73.73 ± 0.00 | 93.34 ± 0.00 | 78.56 ± 0.00 |
DRAN-Net parameter_
| Layer name | Parameters |
| Liner layer | 36×256 |
| Dropout | 0.01 |
| Residual Block1 | 256×256 |
| Attention Residual Block1 | 256×256 |
| Residual Block2 | 256×256 |
| Attention Residual Block2 | 256×256 |
| Residual Block3 | 256×256 |
| Attention Residual Block3 | 256×256 |
| Residual Block4 | 256×256 |
| Dropout | 0.01 |
| Liner layer | 256×30 |
Prediction results of various model methods_
| Model+Representation Set | Accuracy (%) | Top-3 (%) | F1 (%) |
|---|---|---|---|
| Random Forest+T-Pre | 77.39 ± 0.05 | 95.54 ± 0.12 | 80.75 ± 0.12 |
| Gradient Boosting+T-Pre | 76.63 ± 0.21 | 94.58 ± 0.17 | 80.47 ± 0.76 |
| Decision Tree+T-Pre | 72.93 ± 0.10 | 95.17 ± 0.00 | 78.52 ± 0.20 |
| Nearest Neighbor+T-Pre | 72.48 ± 0.00 | 93.68 ± 0.00 | 77.13 ± 0.00 |
| Multilayer Perceptron+T-Pre | 65.35 ± 0.54 | 82.06 ± 1.18 | 58.49 ± 1.57 |
| Support Vector Machine+T-Pre | 62.49 ± 0.00 | 78.70 ± 0.00 | 69.51 ± 0.00 |
| Naive Bayes+T-Pre | 33.60 ± 0.00 | 54.08 ± 0.00 | 29.48 ± 0.00 |
| GraphSAGE+TWIG-Pre | 71.10 ± 1.02 | 95.77 ± 0.88 | 60.41 ± 2.49 |
| GIN+TWIG-Pre | 48.28 ± 0.64 | 84.52 ± 1.57 | 21.82 ± 1.50 |
| GCN+TWIG-Pre | 30.97 ± 0.25 | 56.54 ± 3.23 | 3.83 ± 0.49 |
| Ours (DRAN-Net+T-Pre) | 82.83 ± 1.63 | 98.14 ± 0.32 | 82.80 ± 1.52 |
Statistical significance tests_
| Ours (DRAN-Net+N-Pre) vs. | Metric | t | p-Value (α = 0.05) |
|---|---|---|---|
| Ours (DRAN-Net+N-Pre) vs. Random Forest+T-Pre | Accuracy | 8.4356 | 0.0009 |
| Top-3 | 17.0113 | 0.0000 | |
| F1 | 3.0064 | 0.0391 | |
| Ours vs. Gradient Boosting+T-Pre | Accuracy | 7.4592 | 0.0017 |
| Top-3 | 21.9686 | 0.0000 | |
| F1 | 3.0658 | 0.0154 | |
| Ours vs. Decision Tree+T-Pre | Accuracy | 13.5555 | 0.0002 |
| Top-3 | 20.7535 | 0.0000 | |
| F1 | 6.2425 | 0.0030 | |
| Ours vs. Nearest Neighbor+T-Pre | Accuracy | 14.1983 | 0.0001 |
| Top-3 | 31.1652 | 0.0000 | |
| F1 | 8.3411 | 0.0011 | |
| Ours vs. Multilayer Perceptron+T-Pre | Accuracy | 22.7628 | 0.0000 |
| Top-3 | 29.4089 | 0.0000 | |
| F1 | 24.8754 | 0.0000 | |
| Ours vs. Support Vector Machine+T-Pre | Accuracy | 27.9028 | 0.0000 |
| Top-3 | 135.8411 | 0.0000 | |
| F1 | 19.5509 | 0.0000 | |
| Ours vs. Naive Bayes+T-Pre | Accuracy | 67.5347 | 0.0000 |
| Top-3 | 307.8786 | 0.0000 | |
| F1 | 78.4389 | 0.0000 | |
| Ours vs. GraphSAGE+TWIG-Pre | Accuracy | 13.6408 | 0.0000 |
| Top-3 | 5.6596 | 0.0005 | |
| F1 | 17.1617 | 0.0000 | |
| Ours vs. GIN+TWIG-Pre | Accuracy | 44.1176 | 0.0000 |
| Top-3 | 19.0074 | 0.0000 | |
| F1 | 63.8515 | 0.0000 | |
| Ours vs. GCN+TWIG-Pre | Accuracy | 70.3203 | 0.0000 |
| Top-3 | 28.6586 | 0.0000 | |
| F1 | 110.5693 | 0.0000 |