Real-world time series forecasting project to support airline planning and decision-making.
Airlines rely on demand forecasts to plan capacity, staffing, and budgets.
Inaccurate forecasts can lead to wasted resources or lost revenue.
This project demonstrates how historical passenger data can be used to produce
reliable, business-ready demand forecasts.
- Forecasted future airline passenger demand using historical monthly data
- Compared baseline, statistical, and machine-learning forecasting approaches
- Selected the most reliable model using realistic time-series evaluation
- Delivered results through an interactive Streamlit dashboard
- Used naive and seasonal naive models as baseline benchmarks
- Applied ETS (Holt–Winters) and ARIMA for statistical forecasting
- Built an XGBoost model using lag-based features
- Evaluated models using walk-forward validation to avoid data leakage
- ETS (Holt–Winters) produced the most stable and accurate forecasts
- Final forecasts include confidence intervals to support risk-aware planning
- Results are easy to explore and export via the interactive dashboard
- The model is trained on historical passenger demand from a single airline route or dataset and may not generalize to other airlines, regions, or market conditions.
- External factors such as fuel prices, economic conditions, weather events, and regulatory changes are not explicitly modeled.
- Forecasts assume historical seasonality patterns will continue into the future.
- The machine-learning model relies on lag-based features and may not capture sudden structural breaks (e.g., pandemics or demand shocks) without retraining.
Python, Pandas, NumPy, statsmodels, XGBoost, Streamlit