LNG Trade Macro Risk Assessment and Prediction Based on Artificial Intelligence

Against the backdrop of global energy transition and dramatic shifts in geopolitical landscapes, risk assessment and prediction for liquefied natural gas (LNG) trade have become critical to ensuring energy security. This study develops an LNG trade macro risk dataset encompassing thirteen features categorized into three major domains: political-military factors, economic fundamentals, and social security. Risk levels were collaboratively annotated by domain experts and the DeepSeek large language model. We systematically compared the predictive performance of machine learning models (Support Vector Machines, Random Forest, Naïve Bayes) and deep learning architectures (Multilayer Perceptron, Deep Neural Networks). Experimental results demonstrate that the dual-layer MLP model achieved optimal performance, significantly outperforming both traditional machine learning approaches and deeper DNN configurations. Our findings reveal that controlling network complexity is essential for deep learning models in small-scale data scenarios, as excessive layer depth induces overfitting. This research validates the effectiveness of MLP in LNG trade macro risk prediction and provides reliable technical support for decision-making in energy security management.