NYC Taxi Demand Prediction

Project Overview

This project presents a spatio-temporal forecasting system to predict taxi ride pickups demand within localized coordinates across New York City. The model enables taxi fleets and ride-hailing services to optimize dispatch routes and decrease passenger waiting times.

Key Features & Objectives

• Spatial Segmentation: Groups GPS coordinates into localized pickup regions across NYC.
• Time-Series Tracking: Processes time-aggregated ride metrics to capture temporal demands.
• Inference Regressions: Outlines future demand quantities using historical ride profiles.

Technical Stack & Technologies

• Language: Python
• Machine Learning: Scikit-learn, K-Means Clustering, XGBoost, Random Forests
• Data Analysis: Pandas, NumPy, Matplotlib

Core Techniques & Data Models

• K-Means Spatial Clustering: Segmented NYC coordinates into cluster centroids using K-Means algorithms, transforming continuous coordinate streams into discrete geographic zones.
• Feature Lag Engineering: Engineered temporal lag variables (e.g. pickup counts 10m, 1h, 1d ago) to represent chronological dependencies.
• Regression Modeling: Trained Random Forests and gradient-boosted trees (XGBoost) to predict demand density values for each cluster zone during future time windows.