Every weekday at 5:00 PM, New York City's roads become a battlefield. We analyzed 2 million crashes to find out why.

11 years of collision data from every borough reveals the exact hour, day, and conditions when you're most likely to be hit — and the findings challenge everything the city tells you about traffic safety.

RecordsReveal Data Team May 2026 18 min read 2,018,963 crashes · NYC Open Data

If you live in New York City and drive home at 5:00 PM on a weekday, you are statistically entering the most dangerous hour of your entire week. Not 2:00 AM when the drunk drivers are out. Not rush hour in the morning. 5:00 PM.

We downloaded every motor vehicle collision reported to the NYPD between 2012 and 2023 — all 2,018,963 of them — and ran the data through machine learning models to find patterns invisible to human analysis. What we found should change how every New Yorker thinks about when they commute.

Here's what 11 years of crash data reveals about staying alive on NYC roads.

5PM

Most dangerous hour

Peak injuries at rush hour — when traffic is heaviest and drivers are most impatient

Saturday

Deadliest day of week

More fatal crashes on Saturday than any other day — alcohol and speed are factors

19.9%

Driver distraction

Nearly 1 in 5 crashes attributed to driver inattention — the #1 known cause

Finding #1

The most dangerous hour to drive in NYC is 5:00 PM. The deadliest hour is 4:00 AM.

Total injuries and fatalities by hour of day · 2,018,963 crashes · 2012-2023

This is the finding that should change how you plan your commute. 5:00 PM sees more injuries than any other hour — the evening rush hour when everyone is trying to get home at once, patience is thin, and traffic density peaks.

But the deadliest hour? That's 4:00 AM. Far fewer crashes happen at 4 AM, but when they do, people die. Speed, alcohol, empty roads, and fatigue combine to make early morning the most lethal time on NYC streets.

"The 5 PM spike proves that crash frequency is driven by traffic volume, not driver behavior. More cars = more crashes. It's simple physics."

Finding #2

Saturday is the deadliest day of the week. Friday has the most crashes.

Crashes, injuries, and fatalities by day of week · 2,018,963 crashes

Friday sees the most total crashes — the end-of-week rush as everyone tries to get home or out for the night. But Saturday kills more people than any other day. The pattern is clear: weekend crashes are deadlier because they involve more speed, more alcohol, and more reckless behavior.

Sunday is surprisingly safe. Fewer people on the road, less urgency, and — crucially — fewer drunk drivers than Saturday night.

Finding #3

Brooklyn has the most crashes. Manhattan has the most deaths.

Total injuries and fatalities by NYC borough · 2012-2023

Brooklyn leads in total crash injuries with 142,978 people hurt over 11 years. But Manhattan's roads are deadlier — 312 people killed, the highest fatality count of any borough.

Why? Manhattan's grid layout, higher traffic speeds on avenues, and the sheer density of pedestrians all contribute. Brooklyn has more crashes because it has more cars. Manhattan has more deaths because its crashes are more severe.

Finding #4

Driver distraction causes 1 in 5 crashes. But 34% of crashes have "unspecified" causes.

Top 10 contributing factors · 2,018,963 crashes · NYC collision reports

The single most cited known cause of NYC crashes? Driver inattention/distraction at 19.9%. Phones, passengers, navigation screens — anything that takes your eyes off the road for even two seconds.

But here's the problem: 34.3% of all crashes are listed as "Unspecified." That's 692,913 crashes with no identified cause. Either the reporting is incomplete, or NYPD officers aren't investigating thoroughly enough to assign a factor.

The "Unspecified" problem means we're missing critical information about one-third of all crashes. Better data collection could reveal patterns that save lives.

4 More Things The Data Reveals

Sedans are involved in more crashes than any other vehicle type

No surprise: sedans dominate NYC roads and dominate crash statistics. SUVs are second, followed by taxis and station wagons.

Machine learning can barely predict crash severity

Our best model (Random Forest, R²=0.03) proves that casualty outcomes are nearly random given the data we have. Crashes are fundamentally unpredictable.

Pedestrians account for 18% of injuries but 50% of deaths

111,436 pedestrians injured vs 1,464 killed. When a pedestrian gets hit, they're far more likely to die than a motorist or cyclist.

The data identified 4 distinct crash types via unsupervised learning

K-Means clustering found 4 natural patterns: rush hour injuries, late-night fatalities, weekend high-severity, and midday low-impact crashes.

Methodology

How We Did This

Dataset: NYC Motor Vehicle Collisions — Crashes (NYC Open Data) · 2,018,963 crashes · 2012-2023 · Downloaded via Kaggle.

Supervised Learning: Linear Regression, Ridge, Lasso, and Random Forest trained on 452,283 crashes with casualties. Target: total_casualties. Best model: Random Forest R²=0.0296 (very low — casualties are unpredictable).

Unsupervised Learning: K-Means clustering (K=4) on crash hour, day, borough, and casualties. PCA explains 52% of variance in 2 dimensions.

Tools: Python · pandas · scikit-learn · Plotly · Kaggle API.