NYC Taxi Trip Duration Forecasting at City Scale

Problem

NYC trip duration is driven by time-of-day, origin-destination patterns, and city-scale congestion. Simple time averages mis-price ETAs in operational routing systems, leading to customer dissatisfaction and driver inefficiency.

Approach

Trained XGBoost model on 1.4 million trip records with engineered temporal and geospatial features. Incorporated weather data and traffic patterns using Folium for geographic sanity checks. Implemented Pandas-based data hygiene and leak-safe train-test splits for production credibility.

Result

City-scale ETAs accurate enough for operational routing. Drivers and passengers get reliable time estimates even during peak congestion.

• ◆ETAs accurate enough for operational deployment
• ◆Handles peak congestion, weather, and geospatial patterns
• ◆1.4M trips of real-world training data, not toy notebooks
• ◆Production-ready for routing and driver-ETA systems