Lê Quang Phú
Back to projects
•
4 min read
NYC Taxi Spatiotemporal Analysis

✨ Features

  • Spatiotemporal EDA: Analyzing patterns across time and space with hourly, daily, and monthly views
  • Zone Grouping: Intelligently groups 265 NYC taxi zones into 6 logical regions (Manhattan, Brooklyn, Queens, Bronx, Staten Island, Airports)
  • Behavior-based Anomaly Detection: Identifies unusual trip patterns using heuristic methods
  • Interactive Dashboard: Streamlit-powered visualization with multiple analysis views
  • High-Performance Processing: Uses Polars for efficient data operations on large datasets

🏗️ Architecture

Data Pipeline

  • Data Ingestion: Automated download from NYC TLC API
  • Data Processing: Polars for high-performance DataFrame operations
  • Zone Mapping: 265 taxi zones grouped into 6 regions for spatial analysis

Analysis Modules

  • Temporal Analysis: Time-based pattern discovery (hourly/daily/monthly)
  • Spatial Analysis: Zone-level pickup/dropoff patterns and regional flows
  • Anomaly Detection: Multiple heuristic methods for outlier detection
  • Statistics: Comprehensive summary statistics and aggregations

Dashboard

  • Framework: Streamlit for interactive web-based visualization
  • Visualization: Plotly for interactive charts and graphs
  • Deployment: Streamlit Cloud for zero-cost hosting

⚡ Tech Stack

Core

  • Python 3.9+ - Core development
  • Polars - High-performance DataFrame operations
  • Streamlit - Interactive dashboard framework
  • Plotly - Interactive visualizations

Data Sources

  • NYC TLC API - Taxi trip data (public domain)

🔍 Analysis Capabilities

Temporal Patterns

  • Peak hours identification (5-7 PM evening rush)
  • Day-of-week analysis
  • Monthly trend analysis
  • Weekend vs weekday comparisons

Spatial Patterns

  • Top pickup/dropoff zones by volume
  • Zone-to-zone flow analysis
  • Regional inter-borough flow visualization
  • Zone-hour heatmaps

Anomaly Detection

  • Speed violations (>60 mph)
  • Fare outliers (IQR method)
  • Distance-duration mismatches
  • Late-night high fare anomalies

📊 Key Findings

Temporal Insights

  • Peak activity: 5-7 PM (evening rush hour)
  • Lowest activity: 3-5 AM
  • Weekdays show 15-20% higher activity than weekends

Spatial Insights

  • Manhattan dominates with ~70% of pickup/dropoff activity
  • JFK and LaGuardia show distinct patterns (airport trips)
  • Strong commuter patterns between boroughs

Anomaly Rate

  • ~2-5% of trips flagged as anomalous
  • Late night (10PM-3AM) shows 3x higher anomaly rates

🚀 Live Demo

Visit the interactive dashboard at nyc-taxi-spatiotemporal-analysis.streamlit.app

Select any year and month of available NYC taxi data to explore:

  • Trip patterns by hour, day, and month
  • Top pickup and dropoff zones
  • Regional flow patterns
  • Anomaly detection results

📋 Prerequisites

  • Python 3.9 or higher
  • UV package manager (recommended) or pip

🛠️ Quick Start

1. Clone the Repository

git clone https://github.com/lequangphu/nyc-taxi-spatiotemporal-analysis.git
cd nyc-taxi-spatiotemporal-analysis

2. Install Dependencies

# Using UV (recommended)
uv sync

# Or using pip
pip install -e .

3. Run the Dashboard

streamlit run src/dashboard/app.py

The dashboard will open at http://localhost:8501

📖 Project Structure

nyc-taxi-spatiotemporal-analysis/
├── src/
│   ├── data/
│   │   └── download.py     # Data download from NYC TLC
│   ├── eda/
│   │   ├── spatial.py      # Spatial analysis
│   │   ├── stats.py        # Statistical summaries
│   │   └── temporal.py     # Temporal analysis
│   ├── zones/
│   │   └── grouper.py      # Zone grouping logic
│   ├── anomaly/
│   │   └── detector.py     # Anomaly detection
│   └── dashboard/
│       └── app.py          # Streamlit dashboard
├── data/                   # Data storage
└── pyproject.toml

🎯 CV Keywords Demonstrated

  • Spatiotemporal data analysis
  • Exploratory data analysis (EDA)
  • Time-series pattern analysis
  • Zone/device grouping
  • Behavior-based anomaly detection
  • Python, Polars, SQL
  • Dashboard development (Streamlit)
  • Data visualization (Plotly)