Forecasting air transport demand in Indonesia: A machine learning approach considering socio-economic factors and pandemic impact

Air transport plays a crucial role in Indonesia, the world’s largest archipelago, facilitating economic development and vital connectivity amidst increasing demand. This study addresses the imperative for accurate demand forecasting by developing a machine learning-based model to predict annual air transport passenger numbers in Indonesia. Utilizing annual time series data (1970–2021) from the World Bank, encompassing key macroeconomic and demographic variables. This research employs feature engineering to introduce the ‘Pandemic Trend’ feature and utilizes fuzzy logic for categorical variables representation. Ten regression algorithms were trained and evaluated using a time-based holdout approach, specifically testing generalization capability on the highly abnormal data of 2021. Our results indicate that the Support Vector Regression (SVR) model is the most optimal, demonstrating superior generalization with the lowest Mean Absolute Percentage Error (MAPE = 3.22 %) on the test set. Other complex models, such as XGBoost and Random Forest, suffered from severe overfitting, with MAPEs exceeding 100 %. The validated SVR model was subsequently used to perform medium-term forecasting for 2022, contributing an initial post-pandemic recovery estimate. The findings offer robust tools for policy-making and strategic planning in Indonesia’s aviation sector.