Leveraging Multisource Data with Machine Learning for Earthquake Loss Prediction: A Case Study from Yunnan, China

This study uses Yunnan Province, a seismically active region of China, as a case study to show how well machine learning performs in assessing earthquake risk. Machine learning methods, particularly the XGBoost model, outperform traditional statistical methods in forecasting earthquake-induced economic losses by effectively integrating multisource data sets from seismology, structural engineering, and socioeconomics. An empirical analysis of 108 seismic events (1990–2022) revealed three key insights. First, machine learning improves the accuracy of earthquake loss predictions, and XGBoost has the best performance. Second, the two main factors influencing earthquake-related economic losses are casualties and affected areas, demonstrating that risk exposure is the most important factor. Third, the design of earthquake index insurance should incorporate the degree of risk exposure in addition to the current magnitude-based index. These findings provide a scalable paradigm for applying machine learning in the governance of localized disasters in seismically active developing regions.