This project focuses on forecasting monthly airline passenger traffic using historical data that exhibits clear trend and seasonality patterns.
By leveraging statistical time series models, the goal is to predict future passenger volumes to assist in capacity planning, resource allocation, and strategic decision-making for airline operations.
Three forecasting models are implemented and compared — Holt-Winters Exponential Smoothing, ARIMA, and SARIMAX — to determine which provides the most accurate and stable forecasts.
- Analyze and visualize monthly airline passenger trends and seasonality
- Build and tune Holt-Winters, ARIMA, and SARIMAX forecasting models
- Compare model accuracy and evaluate forecasting performance
- Predict future passenger demand and visualize forecast intervals
- Interpret findings for actionable insights in airline planning
- Time Series Decomposition – To separate trend, seasonality, and residual components
- Stationarity Testing – Using Augmented Dickey-Fuller (ADF) test
- Model Fitting & Parameter Selection – Grid search for ARIMA/SARIMAX hyperparameters
- Forecast Evaluation – Based on metrics such as RMSE, MAPE, and AIC
- Visualization – Forecast plots with confidence intervals to interpret results
Description: Displays the long-term upward trend and recurring seasonal spikes in air travel.
Description: Shows the rolling mean and standard deviation to assess stationarity in the time series.
Description: Breaks the time series into trend, seasonal, and residual components for deeper insight.
Description: Demonstrates how the Holt-Winters model effectively captures both trend and seasonality in the dataset.
Description: Highlights ARIMA’s performance — captures trend but fails to model the seasonal variations accurately.
Description: Shows the SARIMAX model’s superior fit, accurately modeling both seasonal and trend components.
This section summarizes the performance evaluation of the three forecasting models — Holt-Winters, ARIMA, and SARIMAX — applied to the airline passengers dataset.
- Successfully captured both trend and seasonality in the data.
- Forecasted values closely matched the true test data, especially during seasonal peaks.
- Slight underestimation at troughs but strong overall alignment.
- Ideal for datasets with multiplicative seasonality.
- MAE:
10.30 - RMSE:
15.81
✅ Conclusion:
Holt-Winters achieved the lowest error and provided the most accurate forecast.
It serves as the best-performing baseline model for this dataset.
- Accurately modeled the overall upward trend but failed to capture seasonality.
- Forecasted curve appeared too smooth, missing periodic peaks and dips.
- Unsuitable for datasets with strong seasonal patterns.
- MAE:
41.83 - RMSE:
55.22
⚠️ Conclusion:
ARIMA is ineffective for seasonal datasets; forecasts lacked seasonal structure and had high error values.
- Captured both trend and seasonality accurately.
- Forecasts aligned closely with the actual data across the entire test set.
- More robust than ARIMA, allowing inclusion of exogenous regressors.
- MAE:
13.99 - RMSE:
17.20
✅ Conclusion:
SARIMAX produced reliable and stable forecasts, effectively modeling seasonal patterns with slightly higher complexity.
| Model | MAE | RMSE | Key Strengths | Limitations |
|---|---|---|---|---|
| ARIMA | 41.83 | 55.22 | Captures overall trend | Fails to model seasonality |
| Holt-Winters | 10.30 | 15.81 | Best accuracy, captures trend & seasonality | Slight underestimation at troughs |
| SARIMAX | 13.99 | 17.20 | Handles trend, seasonality, and exogenous factors | Slightly higher RMSE than Holt-Winters |
- Best Model: 🥇 Holt-Winters Exponential Smoothing
- Lowest RMSE (
15.81), best alignment with actual values, interpretable, and simple to implement.
- Lowest RMSE (
- Runner-Up: 🥈 SARIMAX
- Nearly as accurate, highly flexible, and suitable for advanced forecasting with external features.
- Least Effective: 🚫 ARIMA
- Poor at modeling seasonal data and resulted in the highest RMSE (
55.22).
- Poor at modeling seasonal data and resulted in the highest RMSE (
Final Takeaway:
For datasets with strong trend and seasonality,
- Use Holt-Winters for simplicity and accuracy.
- Choose SARIMAX for complex scenarios or inclusion of external variables.
- Avoid ARIMA unless data is non-seasonal or stationarized.
- The airline passenger data shows clear seasonality and long-term growth trend.
- SARIMAX and Holt-Winters are both suitable for forecasting, with Holt-Winters slightly outperforming.
- Time series forecasting enables better demand planning and operational forecasting in the airline industry.
- The models provide a framework extendable to similar business forecasting problems.
- Python 3.x
- Pandas, NumPy – Data manipulation and numerical computations
- Matplotlib, Seaborn – Visualization
- Statsmodels – Time series modeling (ARIMA, SARIMAX, Holt-Winters)
- Scikit-learn – Evaluation metrics
- Jupyter Notebook – Interactive development environment
- Clone this repository:
git clone https://github.com/indu-explores-data/Airline-Passenger-Forecasting.git cd Airline-Passenger-Forecasting - Create a virtual environment (Optional):
python -m venv venv source venv/bin/activate # macOS/Linux venv\Scripts\activate # Windows - Install dependencies:
pip install -r requirements.txt - Launch the Jupyter Notebook:
jupyter notebook Airline Passenger Forecasting.ipynb
- Load the dataset (airline-passengers.csv data).
- Run the notebook
Airline Passenger Forecasting.ipynbsequentially to preprocess data, visualize, and fit models. - Evaluate model performance using RMSE and MAE.
- Visualize forecast results and confidence intervals.
- Compare forecasts across ARIMA, Holt-Winters, and SARIMAX.
Let’s connect on LinkedIn for project discussions or data-driven collaborations:
If you found this project helpful, please ⭐ star the repository and share your thoughts. Suggestions and contributions are always welcome!