Freight Rate Dynamics: A Comparative Study of Classical and Machine Learning Models Integrating Macro and Micro Factors

This study analyses the relationships between regional dry bulk freight rates and chartering data enriched with operational, economic, and vessel-specific predictors. The dataset comprises 305 fixture records filtered from Maritime Research and Thomson Reuters Eikon and includes indicators such as ship recycling prices, fuel costs, exchange rates, and global stock indices. Three modeling approaches, Hedonic Regression, Random Forest (RF), and Artificial Neural Networks (ANN), were implemented and evaluated using RMSE, MAE, MAPE, RMSSE, and R2 metrics. Among them, the RF model achieved the highest predictive performance (R2=94.6\%, RMSE=4.45, MAPE=22.81\%). Hedonic Regression yielded the lowest percentage error (MAPE≈12.78%) after log-transforming the dependent variable but had lower explanatory power (R2≈0.70). The ANN model showed moderate accuracy (R2=62.8%). Feature importance and Shapley additive explanations (SHAP) analysis identified cargo size and secondhand vessel prices as key predictors. The findings highlight the suitability of RF for regional freight rate modeling and emphasize the value of machine learning in enhancing transparency and predictive accuracy in maritime economics.