Optimizing Air Travel – Flight Delay Analysis & Prediction

📌 Project Overview

This project analyzes and predicts flight delays using historical airline operational data.
It combines Exploratory Data Analysis (EDA), feature engineering, and machine learning models to:

• Identify patterns in delays
• Understand the main causes
• Predict whether a flight will be delayed
• Estimate delay duration

The project also introduces a custom Operational Adjustability Index (OAI) to prioritize controllable delays for airlines and airports.

---

📂 Data Sources

• Airline_Delay_Cause.csv – Flight delay records and causes
• Download_Column_Definitions.xlsx – Column descriptions

---

🔍 Key Analyses & Findings

Exploratory Data Analysis (EDA)

• Delta Airlines @ ATL had the highest delay counts for a single month
• DFW Airport recorded:
  • Highest weather-related delays in a month
  • Highest security-related delays (possibly due to stricter protocols)
• Seasonality: Minimal; weather-related delays consistent across months

Visuals Produced:

• Top 20 airports by weather, carrier, and security delays
• Delay trends across months
• Delay counts per airline and airport

---

🛠 Preprocessing Steps

1. Missing Values

• ~200–300 NaN values (~0.3% of data)
• Dropped rows with NaNs due to small percentage

2. Feature Engineering

• Carrier-Airport Delay Score: Mean carrier-caused delays per airport

• Weather-Month Score: Mean weather delays per month

• Security-Airport Score: Mean security delays per airport

• Delay Rate: arr_del15 / arr_flights

  • Threshold 0.2 chosen (from histogram distribution) for binary classification target is_delayed

• Operational Adjustability Index (OAI): OAI = 2.0 × Carrier Delay +2.0 × Late Aircraft Delay +1.5 × NAS Delay +1.0 × Weather Delay +0.5 × Security Delay

• Label Encoding for categorical variables:

  • carrier_name → carrier_label
  • airport → airport_label

3. Target Variables

• Classification: is_delayed (binary: 0 or 1)
• Regression: Average arrival delay (avg_arr_delay) for delayed flights

---

📊 Modelling Approach

Classification Models

• Random Forest Classifier – Baseline model
• XGBoost Classifier – Tuned and achieved best performance

Handling Class Imbalance:

• Used scale_pos_weight = (negative / positive) in XGBoost

Performance Metrics:

Model	Accuracy	Precision	Recall	F1 Score
Random Forest	0.84	0.70	0.78	0.74
XGBoost	0.85	0.72	0.80	0.76

---

Regression Models

• Random Forest Regressor – Baseline
• XGBoost Regressor – Lower MAE and better generalization

Best Regression Results (XGBoost):

• Mean Absolute Error (MAE): 4.52 minutes

---

OAI Prediction

• Model: Random Forest Regressor
• Performance:
  • MAE: 333.5 minutes
  • R² Score: 0.996
• Goal: Identify controllable delay factors for operational planning

---

📦 Libraries Used

• Data Processing: pandas, numpy
• Visualization: matplotlib, seaborn
• Machine Learning: scikit-learn, xgboost
• Model Explainability: SHAP (Kernel crash during OAI evaluation)

---

🚀 How to Run

1. Clone the repository:

git clone https://github.com/AggarwalShourya/Optimizing-flight-delay.git
cd Optimizing-flight-delay