NYC Transit Pulse

Predicting NYC public transit delays using machine learning.

Developed in 36 hours at the UCSB Datathon 2025: DataOrbit.

Overview

NYC Transit Pulse is a machine learning-powered system that forecasts delays in New York City’s MTA buses and trains using historical transit data. With both a web interface and REST API, it's designed for commuters, planners, and developers alike.

Features

• Delay prediction for buses
• Time-based predictions using temporal features
• Web interface and REST API endpoints
• Automated data preprocessing pipeline
• Machine learning backend trained on real MTA data

Technologies

• Python 3.8+
• Flask (Web Framework)
• scikit-learn (Machine Learning)
• pandas (Data Processing)
• Kaggle Hub (Dataset Management)

Installation

1. Clone Repository

git clone https://github.com/alexyan1/nyc-transit-delay-predictor.git
cd nyc-transit-delay-predictor

2. Create a virtual environment (optional)

python -m venv venv
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. Run the web app

python app.py

Dataset

Historical MTA delay and schedule data was sourced from:

• Kaggle: New York City Bus Data

Model Details

• Uses Huber Regression
• Time-series features (day, hour, weekday/weekend)
• Evaluation metrics: Accuracy, MAE
• The pre-trained model is in delay_pipeline.pkl
• It can be trained in the file rlm.py

API Endpoints

POST /predict Example:

curl -X POST http://localhost:5000/predict \
     -H "Content-Type: application/json" \
     -d '{
           "stop": "5th Ave",
           "line": "M4",
           "datetime": "2025-04-23T08:30"
         }'

Response:

{
  "stop": "5th Ave",
  "line": "M4",
  "predicted_delay": 5.2,
  "units": "minutes"
}

Screenshots

Contributors

• Alex Yan
• Nathan Ngo