Abstract: We use classification methods from machine learning to predict the destination of global crude oil exports by utilising micro-level crude oil shipment data that incorporates attributes related to the contract, cargo specifications, vessel specifications and macroeconomic conditions. The results show that micro-level information about the oil shipment such as quality and cargo size dominates in the destination prediction. We contribute to the academic literature by providing the first machine learning application to oil shipment data, and by providing new knowledge on the determinants of global crude oil flows. The machine-learning models used to predict the importing country can reach an accuracy of above 71% for the major oil exporting countries based on out-of-sample tests and outperform both naïve models and discrete regression models.

中文翻译：

---

全球石油出口目的地预测：机器学习方法

摘要：我们使用机器学习的分类方法，通过利用包含与合同、货物规格、船舶规格和宏观经济条件相关的属性的微观原油运输数据来预测全球原油出口的目的地。结果表明，关于石油运输的微观信息，如质量和货物大小，在目的地预测中占主导地位。我们通过为石油运输数据提供第一个机器学习应用程序，并通过提供有关全球原油流动决定因素的新知识，为学术文献做出贡献。用于预测进口国的机器学习模型基于样本外测试对主要石油出口国的准确率可以达到 71% 以上，并且优于朴素模型和离散回归模型。