njtracks/penn-station-departures

NJ Transit NY Penn Station Track Assignments

A dataset of train track assignments at New York Penn Station for NJ Transit trains, with historical data starting from April 2023.

Dataset Summary

This dataset contains historical track assignment data for NJ Transit trains departing from New York Penn Station collected during peak evening commute hours. It's designed for machine learning tasks such as predicting which track a train will depart from based on temporal and route features.

This data is critical for solving the "Penn Station Mad Dash", the stressful scramble that occurs when a track is announced only minutes before departure. By analyzing historical patterns in this dataset, developers and data scientists can build models to predict track assignments before they are officially posted, helping commuters position themselves on the correct level and platform ahead of time.

Use Cases:

• Train track prediction models
• Transit pattern analysis
• Time-series forecasting
• Commuter planning tools

Coverage:

• Time Period: April 2023 onwards (optimistically, updated quarterly)
• Records per Quarter: ~4,000-4,500 daily train departures

Dataset Structure

Data Files

The dataset is organized into quarterly Parquet files for efficient loading and incremental updates:

ny_penn_tracks_2023_q2.parquet  (Apr-Jun 2023)
ny_penn_tracks_2023_q3.parquet  (Jul-Sep 2023)
ny_penn_tracks_2023_q4.parquet  (Oct-Dec 2023)
ny_penn_tracks_2024_q1.parquet  (Jan-Mar 2024)
...

Each file contains one record per train per day, representing the final track assignment observed before departure.

Schema

Column	Type	Description
train_id	string	Unique identifier for the train service (e.g., "6742")
train_number	string	Train number (typically same as train_id)
track	string	Target variable - Platform track number (e.g., "1", "14")
observation_time	timestamp	When this track assignment was observed by njtracks.app
departure_time	timestamp	Scheduled departure time of the train
destination	string	Final destination station as reported by NJTransit APIs (e.g., "Trenton &#9992", "Dover -SEC")
line	string	Transit line name (e.g., "Northeast Corridor")
line_abbreviation	string	Line code (e.g., "NEC", "M&E", "NJCL")

Data Filters

This dataset includes only:

• ✅ NJ Transit trains (Amtrak excluded)
• ✅ Trains with assigned tracks (null/empty tracks excluded)
• ✅ NY Penn Station departures only

Usage

Loading the Dataset

from datasets import load_dataset

# Load entire dataset
dataset = load_dataset("njtracks/penn-station-departures")

# Load specific quarters
dataset = load_dataset("njtracks/penn-station-departures",
                       data_files=["ny_penn_tracks_2024_q*.parquet"])

Or with Pandas:

import pandas as pd

df = pd.read_parquet("hf://datasets/njtracks/penn-station-departures/ny_penn_tracks_2024_q1.parquet")

Quick Start Example

import pandas as pd
from datasets import load_dataset

# Load data
dataset = load_dataset("njtracks/penn-station-departures")
df = dataset['train'].to_pandas()

# Add temporal features
df['hour'] = df['departure_time'].dt.hour
df['day_of_week'] = df['departure_time'].dt.dayofweek

# Analyze track patterns
print(df.groupby(['line_abbreviation', 'track']).size())

Data Collection

Methodology

Data is collected from NJ Transit APIs, which provides real-time departure information similar to what's displayed on station monitors. The collection process:

1. Observation Frequency: API polled regularly during peak evening commute periods.
2. Track Resolution: For each train departure, the dataset includes the last observed track assignment before departure
3. Deduplication: One record per train per day (identified by date + train_id)

Data Integrity:

• All timestamps are in Eastern Time (ET)
• Missing or malformed records are excluded
• Track assignments represent actual observed values, not predictions. We make no guarantees about whether trains actually used these platforms or not.

Disclaimer

This is not official NJ Transit data. This dataset was obtained from NJ Transit's public APIs.

• This data is provided "as is" without any warranties regarding accuracy, completeness, or timeliness
• NJ Transit does not endorse or support this dataset
• The information may contain errors, gaps, or outdated information
• For official schedules and real-time information, visit NJTransit.com

Updates

This dataset will, optimistically, receive quarterly updates as new data becomes available.

Metadata

The manifest.json file contains detailed metadata for each quarter:

{
  "quarters": [
    {
      "file": "ny_penn_tracks_2024_q1.parquet",
      "year": 2024,
      "quarter": 1,
      "records": 4481,
      "date_range": {
        "start": "2024-01-01",
        "end": "2024-03-31"
      },
      "unique_trains": 106,
      "unique_tracks": 16
    }
  ]
}

License

This dataset is released under the Open Database License (ODbL) v1.0.

Contact & Contributions

This dataset is maintained by the team at njtracks.app.

• Issues: Report data quality issues or feature requests on the dataset discussion board
• Updates: Watch this dataset for notifications about quarterly updates
• Community: Share your models, analyses, and insights with the community!

---

Built with data from NJ Transit's Developer APIs