This project builds a machine learning model to predict hourly demand for yellow taxis in New York City using historical trip data. It aims to optimize taxi fleet distribution and enhance the passenger experience.
- Predict taxi demand across different NYC regions at 10-minute intervals.
- Support dispatch systems, dynamic pricing, and urban traffic management.
- Deploy a scalable, retrainable ML model ready for production (MLOps).
- Source: NYC Yellow Taxi Trip Data
- Data Range: January 2016 – March 2016
- Records: 35.5 million+
- Features: Pickup/dropoff time and location, trip distance, fare amount, passenger count, etc.
- Languages: Python
- Libraries: Polars, Pandas, NumPy, Scikit-learn, XGBoost, Matplotlib, Seaborn
- Version Control: GitHub
- Notebook Environment: Jupyter Notebooks
- Outlier removal based on NYC bounding box (latitude and longitude).
- Feature engineering: Trip time, trip distance, speed calculation.
- Unix time binning into 10-minute intervals.
- Clustering using MiniBatchKMeans on pickup locations.
- Lag features (previous 5 time intervals of pickups)
- Cluster latitude, longitude, ID
- Weekday encoding
- Exponential moving averages
- Timestamp-based features (10-minute bins)
- Input Features: ft_1, ft_2, ft_3, ft_4, ft_5, latitude, longitude, weekday, exp_avg, cluster_id
- Target Variable: Number of pickups per cluster per 10-minute interval
- Model: Supervised Learning Models (RandomForestRegressor used)
- Mean Absolute Percentage Error (MAPE)
- Mean Squared Error (MSE)
- Simple Flask app (
app.py) and basic frontend (index.html) for predictions.