This study aims to improve the accuracy of agricultural freight cost forecasting by developing and evaluating machine learning models tailored for road transport in grain supply chains. This study addresses the need for robust predictive tools to support cost-efficient logistics decisions in regions that are heavily reliant on road transportation.
Two machine learning techniques were tested: multilayer perceptron (MLP), a type of artificial neural network, and support vector regression (SVR), a supervised learning model for continuous predictions. Both models were trained on a comprehensive dataset of soybean and corn freight records collected in Brazil from 2012 to 2022. To improve the predictive accuracy, the dataset was segmented into short-distance (≤600 km) and long-distance (>600 km) routes. The methodology also included a rigorous data preprocessing phase to address missing values, inconsistencies, and dataset integration issues.
Both models yielded satisfactory forecasting outcomes, with the SVR model demonstrating slight superiority over the MLP model across both distance categories. The use of two specialized models, one for each distance range, resulted in higher accuracy than that of a single general model. The findings also reveal that freight prices behave differently depending on the transport distance, reflecting distinct logistics cost dynamics. These results confirm the potential of machine learning models to support price-behavior forecasting and logistics planning.
This study is limited by its focus on MLP and SVR performance, without explicitly considering contextual factors such as geopolitical events, regulatory changes, weather conditions, or demand-side market fluctuations. It also does not assess the relative importance of each predictor.
Reliable forecasting of road freight prices in grain supply chains can reduce uncertainty for agribusiness actors and support more equitable negotiations among producers, traders, and carriers. By improving cost predictability, machine learning-based models enhance logistics planning and resource allocation, especially in regions facing infrastructure limitations. This contributes to more efficient supply chains, with potential downstream effects on food prices and rural development. Additionally, the findings inform public policy by highlighting the need for targeted infrastructure investments and support for digital innovation in logistics, fostering greater integration between technological advancement and agricultural distribution systems in large, road-dependent economies.
This study advances the field of forecasting methodologies in agricultural logistics by illustrating the benefits of model specialization and meticulous data preparation. It offers one of the initial empirical applications of MLP and SVR techniques for forecasting road freight prices in Brazil's grain supply chains. The proposed framework delivers actionable insights for both public and private stakeholders, thereby enhancing the predictability of transportation costs and improving decision-making processes in the supply chain.
