Optimizing-Air-Travel

✈️ Air Travel Delay Prediction

📌 Project Overview

Flight delays cause major inconvenience to passengers and significant financial costs to airlines. This project focuses on predicting flight delays using historical air travel data, analyzing both occurrence of delays (classification) and delay duration (regression).

The project not only builds accurate predictive models but also uses SHAP explainability to identify controllable factors, ensuring actionable recommendations for airlines.

---

🎯 Objectives

• Analyze historical flight data to uncover delay trends.
• Predict delay occurrence (Yes/No) using classification models.
• Predict delay duration (minutes) using regression models.
• Use SHAP explainability to interpret model predictions.
• Identify controllable vs. external causes of delays.
• Provide operational recommendations to reduce delays.

---

📂 Repository Structure

air-travel-delay-prediction/
├── data/                # Dataset and column definitions
├── notebooks/           # Jupyter Notebook with complete code
├── reports/             # Final report (PDF)
├── requirements.txt     # Project dependencies
├── README.md            # Documentation
└── .gitignore

---

🔎 Data and Preprocessing

• Dataset: Historical flight-level data with multiple features (airline, airport, month, delays, etc.).

• Column definitions: Provided in data/column_definitions.pdf.

• Key preprocessing steps:

  • Removed outcome-leakage features (post-flight delays, cancellations, diversions).

  • Created new pre-flight features:

    • carrier_delay_rate → average delay % per airline
    • airport_delay_rate → average delay % per airport
    • route_delay_rate → delay % for each route
    • season → one-hot encoded seasonal labels

  • Ensured models use only pre-flight features for realistic predictions.

---

🤖 Modeling Approach

Classification (Delay Occurrence)

• Algorithms tested: Logistic Regression, Random Forest, XGBoost

• Best Model: XGBoost

  • Accuracy: 95%
  • F1 Score: 0.97

Regression (Delay Duration in Minutes)

• Algorithm: Gradient Boosting Regressor

• Performance:

  • MAE: ~10.5 minutes
  • RMSE: ~18.5 minutes
  • R² Score: 0.74

---

🔍 Model Interpretability (SHAP)

• Top influential features:

  1. arr_flights (airport arrivals)
  2. route_delay_rate (historical route delay)
  3. season (seasonal trends)
  4. airport_delay_rate, carrier_delay_rate

• Operational Action Index (OAI):

  • 93.91% of model’s weighted focus is on controllable delays.
  • Ensures recommendations are actionable for airline operations.

---

📈 Results & Insights

• Controllable Delay Sources: Late aircraft turnaround, carrier issues (crew, maintenance, gate).

• External Delay Sources: Weather, NAS (airspace), security.

• Actionable Recommendations:

  • Improve turnaround efficiency at key hubs.
  • Enhance crew scheduling and maintenance planning.
  • Smarter routing to avoid congested NAS corridors.
  • Communicate anticipated risks (weather/NAS) proactively to passengers.

---

🚀 How to Run

1. Clone this repository:

   git clone https://github.com/username/air-travel-delay-prediction.git
   cd air-travel-delay-prediction

2. Install dependencies:

   pip install -r requirements.txt

3. Open the notebook:

   jupyter notebook notebooks/flight_delay_notebook.ipynb

---

🛠️ Technologies Used

• Python (Pandas, NumPy, Matplotlib, Seaborn)
• Scikit-learn
• XGBoost
• Gradient Boosting Regressor
• SHAP for interpretability
• Jupyter Notebook

---

📑 Reports

• Detailed methodology and analysis: presentation_deck/Presentation.pdf
• Dataset description: Dataset/Column_definitions.pdf