Deep Graph Networks improve crime hotspot prediction accuracy to 78%

Researchers have developed a new framework using Deep Graph Convolutional Networks (GCNs) to predict crime hotspots. This approach models crime data as a graph, where grid cells are nodes and proximity defines edges, allowing it to capture complex spatial dependencies that traditional methods overlook. Trained on the Chicago Crime Dataset, the multi-layer GCN model achieved 78% classification accuracy and generated interpretable heat maps, demonstrating its effectiveness for predictive policing. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Offers a more accurate method for predictive policing by leveraging graph-based learning to understand spatial crime dependencies.

RANK_REASON This is a research paper detailing a novel framework for crime hotspot prediction using GCNs.

Read on arXiv cs.CL →

• Graph Convolutional Networks
• Chicago Crime Dataset
• KDE
• SVM
• Asifullah Khan

• paper
• other