Forecasting Turkey’s Air Cargo Tonnage: A Comparative Analysis of Statistical Techniques and Machine Learning Methods

With the expanding global economy, the demand for air logistics continues to grow, further emphasizing its significance. However, this increased demand also presents a barrier to the growth of the air transportation sector, which is marked by a high degree of vulnerability. This study aims to forecast cargo volumes in the air logistics sector, which holds considerable growth potential. To achieve this, two statistical models (SARIMA and ARIMAX) and three machine learning methods (Gradient Boosting Regression Tree, Random Forest, and Support Vector Regression) were utilized in a comparative analysis, and forecasts for air cargo volumes were generated using the model with the best performance. The findings reveal that machine learning-based models outperform statistical models when applied to time series data. Specifically, the Random Forest model demonstrated superior performance in forecasting 1-10 month periods, while the Gradient Boosting Regressor (GBR) outperformed other models in 5-month periods. Additionally, the SARIMA model was found to be highly competitive for short-term forecasts. Based on these results, it was determined that the Random Forest model provides higher accuracy for 1-10 month periods, whereas the GBR model excels in 5-month periods. The results further indicate that dynamic modelling strategies achieved through machine learning methods yield more accurate predictions compared to statistical models.