Regression-Trees-on-TLC-Trip-Record-Data

This notebook demonstrates how to build and evaluate a regression tree model to predict taxi tip amounts using the NYC Taxi dataset. Below is a summary of the key steps and findings:

Key Steps

1. Data Preparation:

   • Loaded the dataset containing taxi trip information
   • Examined correlations between features and the target variable (tip_amount)
   • Identified low-correlation features that could potentially be removed

2. Model Training:

   • Split data into training (70%) and testing (30%) sets
   • Built a Decision Tree Regressor with max_depth=8
   • Trained the model on the training data

3. Evaluation:

   • Evaluated model performance using MSE (Mean Squared Error) and R² score
   • MSE: 1.784
   • R²: 0.001 (very low, indicating poor predictive performance)

4. Experimentation:

   • Tested different max_depth values (4, 12)
   • Removed low-correlation features to simplify the model
   • Visualized the decision tree structure

Key Findings

1. Feature Importance:

   • The top 3 features affecting tip amount are:
     1. fare_amount (highest correlation)
     2. tolls_amount
     3. trip_distance

2. Model Performance:

   • The initial model performed poorly (R² ≈ 0)
   • Reducing max_depth to 4 improved performance slightly
   • Increasing max_depth to 12 worsened performance (negative R²), indicating overfitting

3. Feature Selection:

   • Removing low-correlation features (payment_type, VendorID, etc.) had minimal impact on model performance
   • A simplified model using only the top 3 features performed similarly to the full-feature model