Aviation: Flight Delay Prediction System

Overview

This project builds a machine learning binary classification model to predict flight arrival delays (≥15 minutes) using operational flight features.
A Random Forest classifier achieved 78.8% accuracy with strong precision (0.8361) and is deployed as an interactive Streamlit web application.

The model predicts whether a flight will be:

• On-Time (Arrival delay less than 15 minutes)
• Delayed (Arrival delay equal to or greater than 15 minutes)

---

Live Application

🔗 https://flight-delay-analytics.streamlit.app/

---

Problem Statement

Flight delays affect airline efficiency and passenger satisfaction.

Target Variable: is_delayed

• is_delayed = 1 → Arrival delay ≥ 15 minutes
• is_delayed = 0 → Arrival delay < 15 minutes

This project uses Binary Classification to predict the value of is_delayed.

---

Project Structure

.
├── app.py                      # Streamlit application
├── data
│   ├── airlines.csv
│   └── airports.csv
├── models
│   ├── best_model.pkl          # Trained Random Forest model
│   └── scaler.pkl              # StandardScaler parameters
└── notebooks
    └── Aviation_Delay_Analytics.ipynb

---

Features Used

The model uses 10 operational features:

• TAXI_OUT
• WHEELS_OFF
• DEPARTURE_TIME
• TAXI_IN
• SCHEDULED_DEPARTURE
• SCHEDULED_ARRIVAL
• WHEELS_ON
• ARRIVAL_TIME
• ELAPSED_TIME
• MONTH

Ground movement times and schedule timings are the strongest predictors.

---

Methodology

Data Preprocessing

Aviation features exist on different scales (for example, Month = 1–12 vs Time = 0000–2359).
To normalize feature influence, Z-Score Standardization was applied:

$$ z_i = \frac{x_i - \mu_i}{\sigma_i} $$

Where:

• $z_i$ = standardized value of feature $i$
• $x_i$ = original value of feature $i$
• $\mu_i$ = mean of feature $i$ in the training dataset
• $\sigma_i$ = standard deviation of feature $i$ in the training dataset

This ensures all features contribute proportionally to the model.

---

Model Comparison

Model	Accuracy	Precision	Recall	F1-Score
Naive Bayes	0.6264	0.6615	0.5197	0.5821
Logistic Regression	0.7038	0.7330	0.6424	0.6847
KNN	0.6604	0.6806	0.6062	0.6412
SVM	0.6659	0.6959	0.5909	0.6391
Random Forest	0.7882	0.8361	0.7176	0.7723

Selected Model: Random Forest

• Strong overall performance
• High precision (0.8361), which minimizes false delay alerts
• Robust ensemble-based learning

---

Running the Streamlit Application

1️⃣ Create a Python Virtual Environment (Recommended)

python -m venv venv
source venv/bin/activate       # Mac/Linux
venv\Scripts\activate          # Windows

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Launch App

streamlit run app.py

The app will open automatically in your web browser.

---

Test Cases

Test Case 1 — On-Time Scenario

Represents a standard, efficient flight operation.

Field	Input Value
Operating Month	May
Carrier	Delta Air Lines Inc.
Origin Airport	Atlanta International
Scheduled Departure	10:00 (1000)
Actual Departure	10:00 (1000)
Taxi Out	15
Wheels Off	10:15 (1015)
Wheels On	12:00 (1200)
Actual Arrival	12:05 (1205)
Taxi In	5
Total Elapsed Time	125
Scheduled Arrival	12:10 (1210)

Expected Result:
ON-TIME

---

Test Case 2 — High-Probability Delay Scenario

High taxi-out time and significant gap between scheduled and actual departure.

Field	Input Value
Operating Month	December
Carrier	United Air Lines Inc.
Origin Airport	Chicago O'Hare
Scheduled Departure	08:00 (0800)
Actual Departure	09:30 (0930)
Taxi Out	45
Wheels Off	10:15 (1015)
Wheels On	13:00 (1300)
Actual Arrival	13:20 (1320)
Taxi In	20
Total Elapsed Time	230
Scheduled Arrival	11:30 (1130)

Expected Result:
DELAY EXPECTED

---

Dataset

Dataset sourced from:

U.S. Department of Transportation Flight Delays Dataset (Kaggle)
https://www.kaggle.com/datasets/usdot/flight-delays

---

Tech Stack

• Python
• Pandas
• NumPy
• Scikit-Learn
• Streamlit

---

License

For academic and portfolio demonstration purposes.

---

Contribution

If you find this project useful, feel free to star, fork, and contribute.
Pull requests and suggestions for improvement are welcome.