✈️ AeroDelay - Flight Delay Prediction System

A comprehensive machine learning system for predicting flight delays using Random Forest Regression. This project demonstrates end-to-end ML workflow including data preprocessing, feature engineering, model training, hyperparameter tuning, and deployment through a web interface.

🎯 Project Overview

Flight delays cost airlines and passengers billions annually. This project builds a predictive model to forecast flight delays based on historical data, enabling proactive decision-making for airlines, airports, and travelers.

Key Features:

• 🔄 Robust data preprocessing pipeline
• 🎛️ Feature engineering from temporal and categorical data
• 🌲 Random Forest model with hyperparameter tuning
• 📊 Comprehensive model evaluation and visualization
• 🚀 Interactive web application using Streamlit

📁 Project Structure

AeroDelay/
├── data/
│   ├── raw/                    # Original flight data
│   └── processed/              # Preprocessed data ready for modeling
├── models/                     # Saved trained models
├── notebooks/
│   ├── preprocessing.ipynb           # Data exploration and preprocessing
│   └── model_training_and_evaluation.ipynb  # Model training and evaluation
├── src/
│   ├── preprocess.py          # Data preprocessing functions
│   ├── train.py               # Model training pipeline
│   └── predict.py             # Prediction module with FlightDelayPredictor class
├── app.py                     # Streamlit web application
├── requirements.txt           # Project dependencies
└── README.md

🛠️ Installation

Prerequisites

• Python 3.8 or higher
• pip package manager

Setup

1. Clone the repository

git clone <repository-url>
cd AeroDelay

2. Create a virtual environment (recommended)

python -m venv venv
source venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

🚀 Usage

1. Data Preprocessing

Preprocess raw flight data using the preprocessing module:

python src/preprocess.py

Or explore the preprocessing notebook:

jupyter notebook notebooks/preprocessing.ipynb

Preprocessing Steps:

• Remove unnecessary columns (IDs, data leakage features)
• Handle missing values and duplicates
• Feature engineering (flight duration, temporal features, weekend indicator)
• Encode categorical variables (Airline, Origin, Destination, Aircraft Type)
• Scale numeric features using StandardScaler

2. Model Training

Train the Random Forest model with hyperparameter tuning:

python src/train.py

Or use the training notebook:

jupyter notebook notebooks/model_training_and_evaluation.ipynb

Training Process:

• Train/test split (80/20)
• Baseline Random Forest model
• GridSearchCV for hyperparameter optimization
• Model evaluation with multiple metrics
• Feature importance analysis
• Model persistence using joblib

3. Make Predictions

Use the trained model to predict flight delays:

from src.predict import FlightDelayPredictor

# Load model
predictor = FlightDelayPredictor("models/flight_delay_model.pkl")

# Make predictions
predictions = predictor.predict(X_test)

Or run the example:

python src/predict.py

4. Web Application

Launch the interactive Streamlit app:

streamlit run app.py

📊 Top Important Features

1. ScheduledDuration - Flight duration in minutes
2. Distance - Flight distance
3. DepartureHour - Hour of departure
4. Airline_Encoded - Airline carrier
5. Origin_Encoded - Origin airport

🔬 Technical Details

Machine Learning Pipeline

1. Data Preprocessing

   • Label encoding for categorical features
   • Standard scaling for numerical features
   • Time-based feature extraction

2. Model Architecture

   • Algorithm: Random Forest Regressor
   • Hyperparameters: Optimized via GridSearchCV
   • Cross-validation: 3-fold CV

3. Evaluation Metrics

   • Mean Absolute Error (MAE)
   • Root Mean Squared Error (RMSE)
   • Residual analysis

Delay Categories

The system categorizes predictions into severity levels:

• 🟢 On Time: < 15 minutes
• 🟡 Minor Delay: 15-30 minutes
• 🟠 Moderate Delay: 30-60 minutes
• 🔴 Major Delay: > 60 minutes

📈 Visualizations

The notebooks include comprehensive visualizations:

• Distribution of delay minutes
• Average delay by airline
• Feature importance bar charts
• Residual plots
• Predicted vs Actual scatter plots

🧪 Testing the System

Run a quick test to ensure everything works:

# Test preprocessing
python -c "from src.preprocess import preprocess_data; print('✓ Preprocessing module OK')"

# Test prediction
python src/predict.py