Back to the Future: predicting causal relationships influencing oil prices

Jose Rozanec, Beno Šircelj, Michael Cochez, Gregor Leban

Keywords: strategic foresight, graph neural networks, causality mining, applied machine learning

TL;DR: We present a novel approach to strategic foresight leveraging graph neural networks and testing it on a real-world dataset, where we predict causal relationships for 2000 pairs of entities corresponding to events reported between 2022 and 2023.

Abstract: Strategic foresight has been identified as a key tool to enhance policymaking and guide decision-making related to environmental protection. Artificial intelligence (AI) has been scarcely applied to this domain. We present a novel approach to predict causal relationships between entities leveraging graph neural networks (GNN) for link prediction. We test our approach on a real-world dataset, achieving a ROC AUC of 0.8479 when predicting causal relationships over 2000 pairs of entities corresponding to events reported between 2022 and 2023.

Submission Number: 228