An interactive journey through 848,051 crime incidents spanning a decade. Machine learning meets urban safety in this deep investigation of Boston's crime patterns, predictions, and the future of public safety.
"In the data, we find not just numbers, but the heartbeat of a city."
This investigation uses advanced analytics, machine learning, and optimization
algorithms to understand crime patterns in Boston. From the spatial geography
of risk to the temporal rhythms of danger, we reveal what the data tells us
about public safety—and how we can predict and prevent crime before it happens.
Using LightGBM gradient boosting, iGraph network analysis, and OR-Tools
vehicle routing optimization, we've built a comprehensive picture of crime
in one of America's oldest cities.
An introduction to Boston's crime landscape. Understanding the scale, scope, and patterns across 1.98 million incident records from 2015-2025.
Mapping crime across 12 districts. Using iGraph network analysis to reveal spatial relationships and district connectivity patterns.
When does crime strike? Uncovering temporal patterns with cyclical encoding. Peak activity at 3 PM, variations by day of week.
LightGBM DART model predicts violent crime with 0.6497 AUC-ROC. Feature importance reveals spatial features dominate at 49.4%.
OR-Tools VRP solver computes optimal patrol routes. 3 patrol units cover all 12 high-risk districts with 311 total distance units.
Complete technical documentation: GBDT vs DART vs GOSS model comparison, feature engineering pipeline, cyclical encoding for temporal features, and temporal train/test validation strategy.