Скачать или смотреть Detecting Urban Crime Hotspots with DBSCAN, HDBSCAN & K-Means | Harvard Extension Project

This Harvard Extension School project applies geospatial clustering techniques to detect urban crime hotspots using K-Means, DBSCAN, and HDBSCAN.

Built in Python, the system identifies high-risk areas by clustering crime incident data and evaluating model performance with Precision at A% (PAI@A%) and spatial cross-validation.

Key Highlights:
• Achieved PAI@5% = 16.23%, meaning ~16% of future crimes occurred within the top 5% of predicted high-risk zones.
• Demonstrated that DBSCAN and HDBSCAN outperform K-Means at detecting irregular, real-world spatial clusters.
• Visualized results using heatmaps and spatial density plots for urban risk analysis.

Technologies: Python, scikit-learn, GeoPandas, Matplotlib, Seaborn
Metrics: Spatial cross-validation, PAI@A%, precision analysis
Institution: Harvard Extension School

—
This project was developed collaboratively as part of a Harvard Extension School research module.

The presentation highlights how density-based clustering can inform data-driven urban planning and crime risk modeling.