Machine Learning for Airline Ticket Price Prediction

Overview

As part of my Computational Economics coursework, I developed a predictive model to estimate airline ticket prices using a dataset of 140,000 observations with attributes such as departure/arrival airports, dates, airlines, flight duration, stops, cabin class, and distance. The course emphasized working with computer implementations and numerical methods to conduct quantitative research in economics, and this project applied those skills to an airline pricing context.

Approach

• Data Cleaning & Imputation:
  • Removed columns with excessive missing data (e.g., segmentsEquipmentDescription).
  • Imputed missing travel distances using estimated average speed.
  • Filled missing cabin codes using mode imputation and seats remaining using mean imputation.
• Feature Engineering:
  • Converted travelDuration from string format to total minutes.
  • Created new features such as:
    • dayDifference (days between search date and flight date)
    • connections (number of layovers)
    • Encoded cabin codes numerically (coach = 1, premium coach = 2, business = 3, first = 4)
  • Applied log transformations to correct skewed numerical features.
  • Encoded categorical airport codes into integers.
• Modeling:
  • Experimented with Gradient Boosting and Random Forest Regressors.
  • Random Forest achieved better generalization with an R² ≈ 0.66 on the validation split.

Results

Submitted predictions on the unseen test set (final_test.csv) and achieved a top 5 R² score in the class (R² = 0.641).

Key Takeaways

• Strengthened skills in feature engineering and handling messy real-world data.
• Gained experience comparing and tuning ensemble methods (Random Forest, Gradient Boosting).
• Learned how machine learning and numerical methods can be applied to economic data analysis and prediction problems.