A Big Data Approach to Cargo Type Prediction and Its Implications for Oil Trade Estimation

In estimating global crude oil trade flows, existing studies either only consider crude oil tankers or apply external information to distinguish between crude and product oil trips taken by product tankers; these limitations often render the estimation less accurate or lacking replicability. Our methodology directly addresses such issues by applying in a pioneering way the random forest (RF) ensemble learning technique to the Automatic Identification System (AIS) data so as to predict the cargo types of product tankers during their loaded trips. In the process, by leveraging domain knowledge, we propose and construct ten unique input features for the RF model; given the model predictions, we quantify global crude oil trade flows in a more accurate and tractable manner. As an external validation, we apply the resulting shipping dataset and perform a quantitative exercise that studies global crude oil trade fluctuations, which enables relevant stakeholders to quickly identify emerging trade flow risks and better engage in decision-making. Our study further extends existing applications of the AIS data in the domains of operations management and maritime transportation, and facilitates the exploration of the most microscopic characteristics of oil transportation