A Geographically Enhanced LightGBM Approach to Intelligent Freight Transportation Cost Prediction

Transportation cost plays a central role in total logistics costs, yet freight rates vary significantly across different scenarios due to multiple factors such as region, cargo type, and vehicle type. Existing prediction models mostly focus on a single dimension or simple linear relationships, making it difficult to effectively capture geographic heterogeneity, nonlinearity, and multi-factor interactions. To address these issues, this paper constructs four predictive models—Random Forest, CNN, LightGBM, and a geographically enhanced LightGBM by constructing city-level average freight rates as well as inter-city freight rate differences and ratios to explicitly encode spatial dependencies into the model and conducts an in-depth analysis of prediction results from the perspectives of cargo type and vehicle length in Guangxi, China. Experimental results show that the geographically enhanced LightGBM model outperforms the other three benchmark models across key evaluation metrics, verifying the effectiveness of incorporating geographic information for improving freight rate prediction accuracy. This study not only provides multiple model solutions and a multi-dimensional evaluation perspective for national freight rate prediction but also offers practical insights for optimizing logistics resource allocation and formulating differentiated pricing strategies.