<p>Freight rates have long been of great interest to maritime economists and practitioners. In recent years, however, they have become increasingly volatile, with widespread impacts across the maritime value chain. As traditional linear and time series decomposition approaches often struggle with forecasting highly fluctuating freight rates, machine learning (ML) applications are becoming increasingly relevant. Nonetheless, there is no comprehensive study that consolidates both proven ML models and the influential factors used as inputs in these models. Therefore, this study systematically reviews 28 articles published between 2012 and 2024, identifying 17 output (target or dependent) and 59 input (explanatory or independent) variables used in building ML models. Neural Network (NN) architectural framework types of ML models are most commonly employed, although hybrid and specialized models often outperform such standalone approaches. The existing literature has a strong focus on the dry bulk market (61% of studies), where the prevalent data periodicity is weekly (42.9%) or daily (39.3%). Regarding model fit and accuracy assessment, error measurement metrics such as RMSE, MAPE, MAE, and MSE are most frequently used. By consolidating these ML methodologies, error measurement metrics and input factors into a single source, this study maps the state-of-the-art of ML modelling in freight rate forecasting, serving as a valuable resource for advancing predictive accuracy and supporting strategic decision-making in the maritime domain.</p>