Research map / Video / Video segmentation and tracking
Spatial-Temporal Augmentation for Crime Prediction (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes STAug, a spatial-temporal augmented learning framework for crime prediction that combines a Mixup-style CrimeMix augmentation with attention-based spatial-temporal aggregation over region-crime-type graphs.
Problem
Crime prediction on grid-scale regions is hampered by sparse and uncertain historical crime data, which limits model generalization.
Approach
Adapts Mixup to crime tensors (CrimeMix) to synthesize training samples for robustness, and pairs it with multi-head attention over crime types plus graph propagation over spatial and temporal neighbor edges.
Key results
- On NYC-Crimes, STAug reaches micro-F1 0.6573 and macro-F1 0.6567, versus 0.6465 and 0.6413 for the best baseline ST-DPL.
- On Chicago-Crimes, STAug reaches micro-F1 0.7361 and macro-F1 0.7356, versus 0.7237 and 0.7296 for ST-DPL.
- STAug outperforms DeepCrime, ST-SHN, and ST-DPL on both datasets under both metrics.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use STAug instead of ST-DPL | Crime prediction · NYC-Crimes | micro-F1: 0.6465 -> 0.6573 | Helped |
| Use STAug instead of ST-DPL | Crime prediction · NYC-Crimes | macro-F1: 0.6413 -> 0.6567 | Helped |
| Use STAug instead of ST-DPL | Crime prediction · Chicago-Crimes | micro-F1: 0.7237 -> 0.7361 | Helped |
| Use STAug instead of ST-DPL | Crime prediction · Chicago-Crimes | macro-F1: 0.7296 -> 0.7356 | Helped |
Limitations
- Evaluated on only two real-world datasets (NYC and Chicago).
- Student abstract with limited experimental detail; no ablation of CrimeMix reported in the text.
- Hyperparameters, training setup, and hardware are not described.
Experimental setup
Datasets: NYC-Crimes Chicago-Crimes
Benchmarks: NYC-Crimes Chicago-Crimes
Metrics: micro-F1 macro-F1
Baselines: DeepCrime ST-SHN ST-DPL
Builds on: Mixup ST-SHN ST-DPL
Closest papers by meaning
- Multi-Type Urban Crime Prediction (AAAI 2022 · similarity 0.91)
- Hawkes-Enhanced Spatial-Temporal Hypergraph Contrastive Learning Based on Criminal Correlations (AAAI 2024 · similarity 0.87)
- GeoPro-Net: Learning Interpretable Spatiotemporal Prediction Models Through Statistically-Guided Geo-Prototyping (AAAI 2025 · similarity 0.85)
- Towards Dynamic Spatial-Temporal Graph Learning: A Decoupled Perspective (AAAI 2024 · similarity 0.82)
- Get Rid of Isolation: A Continuous Multi-task Spatio-Temporal Learning Framework (NeurIPS 2024 · similarity 0.80)
- MotifGPL: Motif-Enhanced Graph Prototype Learning for Deciphering Urban Social Segregation (AAAI 2025 · similarity 0.80)
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.