Flight delays can significantly disrupt schedules and cause stress for travelers. To address this issue, we have developed a Flight Delay Prediction System that provides real-time and predictive delay estimates based on user input. Travelers can input their flight number, date, and destination to receive:
- Predicted Arrival Delay: Real-time estimates of arrival delays in minutes.
- Flight Reliability Rating: A letter grade (A to E) that summarizes the flight's likelihood of being on time.
Powered by machine learning, this system achieves an R² score of 80%, explaining 80% of the variance in delay times and ensuring predictions deviate from actual delays by no more than 20% on average.
Watch our project in action: Flight Delay Prediction System Demo
- Flight Delay Prediction: Input flight details to predict arrival delay and reliability.
- User-Friendly Interface: Simple and intuitive web-based platform.
- Real-Time and Predictive Insights: Provides accurate delay estimates and grades for better planning.
- High Accuracy: Leveraging XGBoost Regressor with advanced preprocessing for reliable results.
- Input Flight Details: Users enter their flight number, date, departure place, and destination.
- Data Processing: The system matches input data with historical flight records, preprocesses features, and generates predictions using the trained machine learning model.
- Prediction and Rating:
- Predicts arrival delay in minutes.
- Assigns a reliability grade (A to E) based on delay severity.
- Framework: Flask for web application development.
- Machine Learning Models: XGBoost Regressor for delay prediction.
- Data Handling:
pandasfor data manipulation.joblibfor model serialization and loading.
- Dynamic HTML templates to provide an interactive user experience.
- AJAX for real-time data fetching and updates.
- Data Preprocessing:
- Numerical features scaled using
StandardScaler. - Categorical features encoded with
OneHotEncoderandLabelEncoder. - Additional feature engineering for time-based patterns (e.g., sine and cosine transformations of departure times).
- Numerical features scaled using
- Model Training:
- Regression models: Linear Regression, Random Forest Regressor, XGBoost Regressor.
- Model comparison based on MAE, MSE, RMSE, and R² scores.
- Cross-validation to ensure robustness and generalization.
- Python 3.8 or higher
pipfor dependency management
-
Clone this repository:
git clone https://github.com/your-repo/flight-delay-prediction.git cd flight-delay-prediction -
Install dependencies:
pip install -r requirements.txt
-
Download required datasets and place them in the
./static/data/folder:flights.csvDeparturePlace.csvArrivalPlace.csv
-
Ensure trained models (
XGBRegressor_AD_model.pkl,label_encoder.pkl) are stored in./static/models/.
- Start the Flask server:
python app.py
- Open your web browser and navigate to:
http://127.0.0.1:5000/
- Enter flight details:
- Flight number:
UA123 - Departure date:
2024-01-15 - Departure place:
San Francisco - Destination:
Los Angeles
- Flight number:
- Submit the form and view predictions:
- Predicted Arrival Delay:
10 minutes - Reliability Rating:
B
- Predicted Arrival Delay:
We compared three models during development:
| Model | MAE | MSE | RMSE | R² |
|---|---|---|---|---|
| Linear Regression | 5.86 | 56.69 | 7.53 | 0.79 |
| Random Forest Regressor | 6.13 | 61.19 | 7.82 | 0.78 |
| XGBoost Regressor | 5.80 | 55.92 | 7.48 | 0.80 |
XGBoost Regressor was selected due to its superior performance across all metrics.
The learning curve shows that the XGBoost model achieves a balanced trade-off between bias and variance, ensuring robust predictions.
- Hyperparameter Optimization: Implement grid search or random search for model tuning.
- Feature Importance Analysis: Visualize key predictors of flight delays.
- Enhanced Data Sources: Integrate weather and airport traffic data for improved predictions.