Taxi Fare Prediction Model

Introduction

New York City TLC is an agency responsible for licensing and regulating New York City's taxi cabs and for-hire vehicles. The agency wants to develop a regression model that helps estimate taxi fares before the ride, based on the data that TLC has gathered. The TLC data comes from over 200,000 taxi and limousine licensees, making approximately one million combined trips per day.

After developing the regression model and analyzing insights from the A/B test, their curiosity grew about creating a predictive model to identify customers likely to tip. This could help drivers increase revenue and improve satisfaction. 

All the files related to this project are available at Github.com/nitin6753/Fare_Prediction_NYC-TLC

Summary

Objective: 

• Develop a regression model to estimate taxi fares before the ride using NYC TLC data.

• Develop a classification model to predict a customer would tip generously.

Tasks Performed:

• Conducted EDA on 18 variables and identified key variables: trip distance, duration, and fare amount

• Analyzed ride patterns, fare distributions, finding higher tips with credit card payments (A/B test).

• Performed Feature Engineering on dataset:

  • • Grouped dataset based on pick up and drop off locations, to obtain mean predictor variable

    • Created a rush hour variable based on pickup times in morning and evening.

• Developed a Multiple Linear Regression model predicting fares, with mean distance, mean duration, no. of passengers, vendor ID and rush hour as predictor variables.

  • • Developed a Random Forest and Gradient Boosting model to identify if a customer would tip more than 20% of the fare price or not.

    • Feature Engineered on dataset:

      • • Broke the pickup-time into 4 time slots.

        • Used Predicted fare prices, mean duration, and mean distance.

        • Performed hyperparameter tuning and cross validation for both models using GridSearchCV.

• Provided a detailed executive summary and visualizations to stakeholders.

Key Result Areas:

• Promoting credit card payments could increase driver revenue as they are tipped more.

• Achieved strong model performance (R²: 86.4%, MAE: 1.91, RMSE: 3.13).

• Gradient Boosting model was considered the champion model that performed well on test datasets (F1: 74.76%, Recall: 80.71%, Precision: 69.62%, Accuracy: 71.31%).

• According to this model Vendor, predicted fare, mean distance, and mean duration are the most important features.

Recommended: 

App for fare prediction could be developed to enhance user experience and driver planning.

Skills Demonstrated:

EDA, A/B testing, Feature Engineering, Regression Analysis, Hypothesis Testing, Machine Learning, Random Forest, Gradient Boosting, Python, Pandas, Tableau.