Research map / Graph learning / GNN architectures
Urban Incident Prediction with Graph Neural Networks: Integrating Government Ratings and Crowdsourced Reports
Sidhika Balachandar, Shuvom Sadhuka, Bonnie Berger, Emma Pierson, Nikhil Garg
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add rating data (inspection ratings) to training alongside report data (ablation) | URBAN · NYC incidents (real data) | correlation of predicted vs true ratings: 0.10 -> 0.53 | Improved |
| add rating data to training alongside report data (ablation) | URBAN · NYC incidents (semi-synthetic) | correlation of predicted vs true ratings: 0.33 -> 0.62 | Improved |
| add report data to training alongside rating data (full vs ratings-only) (ablation) | URBAN · NYC incidents (real data, full ratings) | correlation of predicted vs true ratings: 0.5223 -> 0.5303 | No clear effect |
| add report data to training alongside rating data (full vs ratings-only) (ablation) | URBAN · NYC incidents (semi-synthetic, full ratings) | correlation of predicted vs true ratings: 0.6246 -> 0.6243 | No clear effect |
| add report data when ratings are artificially subsampled (rodent type) (ablation) | URBAN · NYC incidents (real data, rodent, subsampled ratings) | correlation of predicted vs true ratings: ratings-only model approaches random prediction (r=0) while full model stays correlated | Improved |
| use full URBAN model instead of reports-only for types with unobserved ratings (rarest types) (ablation) | URBAN · NYC incidents (semi-synthetic, type frequency 0-20%) | correlation of predicted vs true ratings: -0.074 -> 0.269 | Improved |
| use full URBAN model instead of reports-only for most frequent types with unobserved ratings (ablation) | URBAN · NYC incidents (semi-synthetic, type frequency 80-100%) | correlation of predicted vs true ratings: 0.745 -> 0.768 | Improved |
| replace flexible multi-type GNN reports-only model by per-type Gaussian process on reports | reports-only baseline · NYC incidents (real data) | correlation of predicted vs true ratings: 0.10 -> 0.03 | Hurt |
| add rating data when predicting report probabilities (full vs reports-only) (ablation) | URBAN · NYC incidents (real data) | correlation of predicted vs true report probability: 0.5406 -> 0.2371 | Hurt |
| add rating data when predicting report probabilities (full vs reports-only) (ablation) | URBAN · NYC incidents (semi-synthetic) | correlation of predicted vs true report probability: 0.4840 -> 0.4680 | No clear effect |
| model higher median income in a tract | URBAN reporting model · NYC incidents (real data) | reporting coefficient: 0.173 ± 0.020 | Improved |
Closest papers by meaning
- TraffiDent: A Dataset for Understanding the Interplay Between Traffic Dynamics and Incidents (NeurIPS 2025 · similarity 0.86)
- GeoPro-Net: Learning Interpretable Spatiotemporal Prediction Models Through Statistically-Guided Geo-Prototyping (AAAI 2025 · similarity 0.85)
- A Bayesian Spatial Model to Correct Under-Reporting in Urban Crowdsourcing (AAAI 2024 · similarity 0.84)
- Graph Mixture of Experts and Memory-augmented Routers for Multivariate Time Series Anomaly Detection (AAAI 2025 · similarity 0.84)
- Spatial-Temporal Augmentation for Crime Prediction (Student Abstract) (AAAI 2024 · similarity 0.82)
- Spatio-Temporal Pivotal Graph Neural Networks for Traffic Flow Forecasting (AAAI 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
