Figure 1.
Literature search scheme (inspired by the Prisma searching model)
Source: own elaboration
Figure 2.
Distribution of types of methods appearing in selected publications
Source: own elaboration
Appendix 1.
Information about most important characteristics of the research undertaken concerning road network selection.
Selected references considered in this paper are included, as many of the works lack quantitative results’ evaluation
| Author | Year | Approach | Source scale | Target scale | Method | Major metric | Major purpose |
|---|---|---|---|---|---|---|---|
| Xiao et al. | 2024 | ML | 1:100 000 | 1:200 000 | Accuracy | database generalization | |
| MLSU-TAGCN | 81.40% | ||||||
| MLSU-GCN | 74.80% | ||||||
| MLSU-GAT | 75.51% | ||||||
| MLSU-GraphSAGE | 80.80% | ||||||
| Selection-4Fs | 71.20% | ||||||
| Selection-22Fs | 78.40% | ||||||
| Tang et al. | 2024 | ML | 1:10 000 | F1-score | database generalization | ||
| 1:50 000 | GCN with functional semantic features | 89.74% | |||||
| 1:200 000 | 82.70% | ||||||
| Zheng et al. | 2024 | ML | 1:250 000 | 1:1 000 000 | Accuracy | database generalization | |
| HAN | 75.35% | ||||||
| Karsznia et al. | 2024a | ML | 1:250 000 | 1:500 000 | Accuracy | general geographic map | |
| DT | 81.25% | ||||||
| RF | 84.38% | ||||||
| SVM | 84.38% | ||||||
| DTGA | 90.00% | ||||||
| NN | 81.88% | ||||||
| Guo et al. | 2023 | ML | 1:10 000 | 1:50 000 | F1 Score | database generalization | |
| GNN | 92.10% | ||||||
| AHP | 88.00% | ||||||
| Lyu et al. | 2022 | Graph | 1:50 000 | 1:100 000 | Accuracy | database generalization | |
| Road-path selection constrained by settlements | 86.00% | ||||||
| Pung et al. | 2022 | Graph | 1:10 000 | „Large-scale” | Selected-Source Correlation | urban road generalization | |
| Functional node elimination | Pearson (ρ) = 0.964 | ||||||
| Spearman (R) = 0.911 | |||||||
| Karsznia et al. | 2022 | ML | 1:250 000 | Accuracy | database generalization | ||
| 1:500 000 | DT | 84.46% | |||||
| DTGA | 83.33% | ||||||
| RF | 84.96% | ||||||
| 1:1 000 000 | DT | 99.18% | |||||
| DTGA | 99.44% | ||||||
| RF | 99.34% | ||||||
| Wu et al. | 2022 | Mesh | 1:10 000 | 1:50 000 | Shape similarity overlap | topographic map | |
| Direct pair merging | 100% | ||||||
| Iterative area elimination | 100% | ||||||
| Zheng et al. | 2021 | ML | 1:10 000 | 1:100 000 | Accuracy | database generalization | |
| MLP | 85.83% | ||||||
| JK-GAT | 88.12% | ||||||
| Res-GAT | 87.88% | ||||||
| Dense-GAT | 87.41% | ||||||
| Han et al. | 2020 | Stroke | 1:5 000 | 1:200 000 | Common stroke ratio | database generalization | |
| AHP | 89% | ||||||
| Yu et al. | 2020 | Stroke | Unknown | Maximum similarity | navigation | ||
| 1:5 000 | Traffic Flow Radical Law Strokes | 61.15% | |||||
| 1:25 000 | 65.86% | ||||||
| 1:50 000 | 90.95% | ||||||
| 1:5 000 | Traffic Flow Pair Strokes | 61.61% | |||||
| 1:25 000 | 65.58% | ||||||
| 1:50 000 | 90.95% | ||||||
| Li et al. | 2020 | Mesh, Stroke | 1:10 000 | 1:50 000 | Maximum similarity | topographic map | |
| Mesh elimination | 89.52% | ||||||
| Stroke-edge elimination | 91.64% | ||||||
| Park, Huh | 2019 | ML | 1:5 000 | 1:25 000 | Matching ratio | topographic map | |
| Logistic Regression | 81.66% | ||||||
| Zhang et al. | 2017 | Stroke | 1:10 000 | 1:50 000 | Accuracy | database generalization | |
| Stroke generation with weighted Voronoi diagrams | 88.80% | ||||||
| Zhou, Li | 2017 | ML | Accuracy | database generalization | |||
| 1:20 000 | 1:50 000 | MP | 80.45% | ||||
| SVM | 77.05% | ||||||
| BLR | 80.90% | ||||||
| 1:100 000 | MP | 79.90% | |||||
| SVM | 81.10% | ||||||
| BLR | 80.65% | ||||||
| 1:200 000 | MP | 91.55% | |||||
| SVM | 92.90% | ||||||
| BLR | 92.35% | ||||||
| 1:50 000 | 1:250 000 | MP | 83.20% | ||||
| SVM | 82.70% | ||||||
| BLR | 83.10% | ||||||
| Weiss, Weibel | 2014 | Stroke | 1:10 000 | 1:200 000 | Mean improvement (vs. Basic) | database generalization | |
| Enhanced stroke generation | 67.88% | ||||||
| Benz, Weibel | 2014 | Stroke, Mesh | 1:10 000 | 1:50 000 | Satisfaction of hard constraints | database generalization | |
| Extended stroke–mesh combination | 100% | ||||||
| Zhou, Li | 2014 | ML | 1:20 000 | Accuracy | map updates | ||
| 1:50 000 | BPNN | 82.4% | |||||
| 1:100 000 | 87% | ||||||
| 1: 200 000 | 98.6% | ||||||
| Li et al. | 2012 | Stroke, Mesh | Accuracy | database generalization | |||
| 1:20 000 | 1:50 000 | Stroke generation | 84.7% | ||||
| 1:100 000 | 76.7% | ||||||
| 1:200 000 | 68.5% | ||||||
| 1:50 000 | 1:250 000 | 77.3% | |||||
| 1:20 000 | 1:50 000 | Mesh density | 67.7% | ||||
| 1:100 000 | 63.7% | ||||||
| 1:200 000 | 61.6% | ||||||
| 1:50 000 | 1:250 000 | 71.4% | |||||
| Zhang, Li | 2011 | Graph | No. of road segments | navigation | |||
| Scale free | Top 2% strokes | Ego network | 82.1% | ||||
| Top 10% strokes | 89.9% | ||||||
| Top 15% strokes | 92.5% | ||||||
| Top 20% strokes | 92.6% | ||||||
| Top 2% strokes | Weighted ego network | 87.1% | |||||
| Top 10% strokes | 95.8% | ||||||
| Top 15% strokes | 94% | ||||||
| Top 20% strokes | 94.6% | ||||||
| Gülgen, Gökgöz | 2011 | Mesh | Selected road length change | database generalization | |||
| 1:25 000 | 1:50 000 | Urban block amalgamation | 14.10% | ||||
| 1:100 000 | −17.30% | ||||||
| Yang et al. | 2011 | Stroke | 1:1 000 | Not specified | Mean similarity with target | database generalization | |
| Hierarchical stroke generation | 40.75% | ||||||
| Touya | 2010 | Stroke, Mesh | 1:50 000 | 1:100 000 | Road length overlap | database generalization | |
| Enriched structural selection | 97% | ||||||
| Liu et al. | 2010 | Stroke | 1:10 000 | 1:50 000 | Avg no of identical strokes | database generalization | |
| Stroke generation with seed extension | 91.84% | ||||||
| Chen et al. | 2009 | Mesh | 1:10 000 | 1:50 000 | Mean consistency with existing map | map updates | |
| Mesh density-based selection | 89.50% |