NYC Taxi Fare Prediction

A neural network model for predicting New York City taxi fares using geospatial and temporal feature engineering.

Built as a final project for CAP4770 (Introduction to Data Science), this project uses the Kaggle NYC Taxi Fare Prediction dataset to achieve a validation MAE of ~$1.98.

Features

• Data Pipeline: Cleans outliers, validates coordinates, and filters trips outside NYC bounding box
• Feature Engineering: Haversine distance calculation, KMeans location clustering (5 zones), peak hour detection
• Visualization: Interactive Folium maps, density heatmaps, temporal trend analysis
• Neural Network: TensorFlow/Keras model with early stopping and adaptive learning rate

Tech Stack

• Data: Pandas, NumPy
• Visualization: Matplotlib, Seaborn, Folium
• ML: scikit-learn (preprocessing), TensorFlow/Keras (model)

Getting Started

Prerequisites

• Python 3.x
• Jupyter Notebook or JupyterLab

Installation

pip install pandas numpy matplotlib seaborn folium scikit-learn tensorflow

Usage

1. Download training data from Kaggle
2. Place train.csv in ./new-york-city-taxi-fare-prediction/
3. Run the notebook:

jupyter notebook Main.ipynb

Model Architecture

Input (17 features) --> Dense(128, ReLU) --> Dense(64, ReLU) --> Output(1)

Features used: distance, pickup/dropoff clusters, hour, weekday (one-hot), month (one-hot), is_peak_hour, passenger_count

Training: Adam optimizer, MAE loss, early stopping (patience=5), learning rate reduction on plateau

Results

Metric	Value
Validation MAE	$1.98
Training Samples	~800K (after cleaning)
Test Split	20%

Project Structure

CAP4770-Final_Project/
├── Main.ipynb                              # Main analysis notebook
├── new-york-city-taxi-fare-prediction/     # Data directory (gitignored)
│   └── train.csv                           # Kaggle dataset
├── .portfolio/                             # Project documentation
│   ├── architecture.md                     # System design and diagrams
│   ├── stack.md                            # Technology decisions
│   └── qa.md                               # Q&A knowledge base
└── README.md

Visualizations

The notebook includes:

• Pickup/dropoff density heatmaps
• Interactive cluster maps with Folium
• Fare trends by hour and day of week
• Distance vs fare scatter plots
• Correlation heatmaps
• Monthly fare trends by year

Author

Jacob Kanfer - GitHub