NYC Taxi Trip Duration Forecasting with Stacked Regression

This project predicts NYC taxi trip durations using stacked ensemble regression models. It includes end-to-end steps from data cleaning, outlier treatment, distance calculation, feature engineering, model selection, cross-validation, and performance evaluation.

---

Features

• Reads and processes NYC taxi dataset (nyc_taxi_trip_duration.csv)
• Calculates distance using geopy (geodesic) from coordinates
• Outlier detection and removal for both trip duration and distance
• One-hot encoding and correlation analysis
• Trains multiple regression models:
  • K-Nearest Neighbors (KNR)
  • Stochastic Gradient Descent (SGDR)
  • Gradient Boosting Regressor (GBR)
  • Bayesian Ridge (BR)
  • Linear Regression (LR)
• Hyperparameter tuning with GridSearchCV
• Stacked regressors with varying final estimators (KNR, GBR, LR)
• Performance metrics:
  • R² Score
  • Mean Absolute Error (MAE)
  • Computation Time

---

Visualizations

• Box plots for outlier detection
• Scatter plots of Distance vs Trip Duration
• Heatmap of feature correlations

---

How to Run

1. Upload nyc_taxi_trip_duration.csv via Google Colab or local Jupyter
2. Run the notebook from top to bottom
3. The models train and output results automatically

---

Requirements

See requirements.txt

---

Author

Akhdan Arifuddin