The application of machine learning (ML) for the automatic classification of building elements is a powerful technique for ensuring information integrity in building information models (BIMs). Previous work has demonstrated the favorable performance of such models on classification tasks using geometric information. This research explores the hypothesis that incorporating contextual information into the ML models can improve classification accuracy. To test this, we created a graph data structure where each building element is represented as a node assigned with basic geometric information. The connections between the graph nodes (edges) represent the immediate neighbors of that node, capturing the contextual information expressed in the BIM model. We devised a process for extracting graphs from BIM files and used it to construct a graph dataset of over 42,000 building elements and used the data to train several types of ML models.
| Authors | Austern, G.; Capeluto, I. G.; Hassoun, S. |
|---|---|
| Year | 2024 |
| Venue | Buildings 14, 527 |
| DOI | 10.3390/buildings14020527 |
| Source Database | Backtracking |
| Bridge-to-GNN | Category A |
| GNN Architecture | GraphSAGE (heuristic — verify) |
| Graph Encoding | BIM Element Topology Graph (inferred from title) |
| AEC Task | Element Classification (inferred from title) |
| Cohort | Mature Applications (2024–2026) |
| Dataset Size | 22 buildings (heuristic — verify) |
|---|---|
| Implementation Framework | PyTorch Geometric (PyG), PyTorch, scikit-learn, NetworkX (heuristic — verify) |
| Key Hyperparameters | 200 epochs; lr 0.01; optimizer=Adam optimizer; 70 hidden units (heuristic — verify) |
| Primary Metric | Accuracy (heuristic — verify) |
| Primary Value | 85% (heuristic — verify) |
| Key Finding | novel approach to enhance this accuracy by incorporating contextual information using geometric proximity graphs (GPGs) to represent building elements (heuristic — verify) |
| Quality Assessment | 3/6 · rigor: Medium · Benchmark: ✓Baseline: ✓CV/Split: ✓ (heuristic — verify) |
| GNN architectures detected | GraphSAGE / SAGEConv · Graph Transformer / GraphGPS / SAN · Knowledge Graph Embedding (TransE/RotatE/DistMult) · GCN |
|---|---|
| Frameworks / libraries | PyTorch Geometric (PyG) · PyTorch · scikit-learn · NetworkX · Topologic / Topologicpy · Grasshopper / Rhino · Revit / Dynamo · Neo4j / Cypher · RDF / OWL / SPARQL |
| Benchmark datasets referenced | PubMed · PROTEINS |
| Hyperparameters (regex-detected) | epochs=200 · learning_rate=0.01 · hidden_units=70 · optimizer=Adam optimizer |
| Dataset stats found | 22 buildings · 907 buildings |
| Reported metrics + values | Accuracy: 79% · Accuracy: 89.9% · Accuracy: 85% · Accuracy: 86.7% · Accuracy: 95% · Accuracy: 99% |
| Quality (heuristic, 6-flag) | 4/6 · rigor: Medium Code/Data: —Benchmark: ✓Baseline: ✓CV/Split: ✓Ablation: —Reproducible: ✓ |
Part of the GML/GNN in AEC Systematic Review (PRISMA 2020) — 112 papers, 2020–2026