Resilience in the skies: Modeling domestic and international air passenger demand in multiple Indian airports

This study presents a comparative analysis of univariate time-series (ARIMA, SARIMA, ETS) and deep learning models (RNN and LSTM) for forecasting post-COVID domestic and international air passenger demand at eight major Indian airports: Ahmedabad, Bengaluru, Mumbai, Kolkata, Delhi, Hyderabad, Chennai, and Pune. Utilizing quarterly data from 2016 to 2023, performance of time-series and deep learning models is evaluated against actual 2024 air traffic data using MAPE, MAE, and RMSE indices. Results demonstrate that model efficacy is highly context-specific. SARIMA consistently outperforms ARIMA in capturing seasonality, while LSTM excels at modeling non-linear complexities, and ETS proves robust for airports with clear trends. Crucially, a SARIMAX model integrating exogenous drivers, including net domestic product, network connectivity, and operational metrics, significantly enhanced forecasting accuracy, particularly for international travel, underscoring the importance of these drivers. The coefficients reveal several interesting policy scenarios, such as enhancing domestic and international connectivity, particularly at emerging hubs, stimulates passenger growth, while densely populated catchments require investments in multimodal integration to counter negative demand. The findings challenge the presumption of a universal forecasting framework and underscore the inefficiency of relying solely on univariate models, advocating for a tailored approach that incorporates key exogenous variables for resilient air traffic management.