Six Years of Subway Crime: What 99,000 Incidents Actually Tell Us
The headline: Trakkr's safety forecast catches roughly 97 out of every 100 stations that turn into safety hotspots — built from 99,462 publicly available NYPD records, weighted by how serious each incident was, and adjusted for how busy each station is.
Raw crime counts are the wrong number. They've always been the wrong number, and every transit crime map you've ever seen is built on them anyway.
Here's the problem: a station that processes 60,000 riders a day and logs 300 incidents a year isn't more dangerous than a station with 800 riders a day and 30 incidents. Divide by ridership and the picture inverts. The "safe" high-volume station stays safe. The low-volume station with a fifth of the incident count becomes the actual high-risk location.
How TrakkRecord™ builds the crime signal
Starting from 99,462 NYPD transit incident records across 2020–Q1 2026, we applied three adjustments before a single model was trained:
Robbery, assault, grand larceny — crimes that pose direct physical risk to riders. Counted at 3× in the weighted score.
Petit larceny, harassment, criminal mischief. Meaningful but lower-severity than felonies — counted at 2×.
Fare evasion, disorderly conduct. Still part of the pattern but appropriately down-weighted against felony incidents.
After weighting, each station's monthly score is divided by average monthly ridership — sourced from MTA Hourly Ridership data (11 months averaged, 71.7% station-level match, remainder filled with borough averages). The result is a crime density score per 1,000 riders.
Direction matters as much as level
A station that's currently a moderate-risk place but was a low-risk place three months ago is more concerning than one that's been steady-state risky for two years. Trakkr tracks not just where a station stands, but which direction it's moving.
By combining recent month-over-month changes with one-, three-, and six-month rolling averages, Trakkr sees both the current temperature and the trajectory. That's what separates it from a static crime map you'd find on any city portal.
A station is "in trouble" if it ranks top-5 in its own borough
Trakkr flags a station as a hotspot if it lands in the top five most dangerous in its own borough for that month. The Bronx is compared to other Bronx stations. Manhattan to Manhattan. Each borough has very different baselines, and ranking them on one giant list would be misleading: Bronx stations would always look "safer" than they are because Manhattan's totals are bigger overall.
The point is to find the stations where the risk is genuinely elevated for the people who use them — not the ones that just look bad because they're crowded.