Customer Booking Prediction Analysis

Project Overview

This project focuses on analyzing and predicting customer booking completion using a machine learning model. The dataset contains booking information with various features related to customer behavior and flight details.

Problem Statement

The goal is to predict whether a customer will complete their booking using relevant features. The dataset is highly imbalanced, with approximately 85% of cases where customers do not complete their booking.

Key Insights

• Target Variable Distribution:
  • 85% of bookings are incomplete (booking_complete = 0).
  • Only 15% of customers complete their bookings.
• Feature Importance:
  • Significant Features: Purchase Lead, Route, Flight Hour, Length of Stay, Booking Origin
  • Passenger preferences like extra baggage, preferred seat, and in-flight meals have lesser importance.
  • Strong interaction effect between Num Passengers and Sales Channel.
• SHAP Interaction Effects:
  • Average Purchase Lead: 84.94 days before flight.
  • Average Length of Stay: 23.04 days.
  • Most flights are scheduled around 9 AM.
  • 66.9% passengers carried extra baggage.

Model Performance

• Accuracy: 85.15%
• Classification Report:
  • Precision for booking_complete = 1: 44%
  • Recall for booking_complete = 1: 7%
• Confusion Matrix:
  • True Negatives: 1682
  • False Positives: 27
  • False Negatives: 270
  • True Positives: 21

Tools & Technologies Used

• Python for data analysis and modeling
• Pandas and NumPy for data manipulation
• Matplotlib and Seaborn for data visualization
• Scikit-Learn for model building
• SHAP for model interpretability

Next Steps

• Address class imbalance using SMOTE or other resampling techniques.
• Experiment with alternative models (e.g., XGBoost, SVM).
• Improve feature engineering.