Clustering analysis and artificial intelligence forecasting of seismic activity in the Sichuan-Yunnan region

This study focuses on the Sichuan-Yunnan region of China, a seismically active area in which major earthquakes (M ≥ 6) have long recurrence intervals and limited observational samples, posing significant challenges for earthquake prediction.We focus on 14 major active fault zones in this region and utilize the shortest distance from earthquakes to faults to explore their relationships, then applied Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to further analyze spatial cluster characteristics between earthquakes to enhance spatial resolution. By combining two spatial partitioning methods, focal mechanism solutions and affiliation ratios, successfully assigned seismic events to their respective fault zones. The analysis highlights fault zones which is more active in M≥6 earthquakes, specifically the Longmenshan Fault Zone, Xianshuihe Fault Zone, Xiaojinhe Fault Zone, Honghe Fault Zone, Lancangjiang Fault Zone and Nujiang Fault Zone. A sliding window approach is applied to calculate 11 feature factors in the time window, and the Synthetic Minority Oversampling Technique (SMOTE) algorithm is implemented to oversample features related to M≥6 earthquakes, artificially augmenting the major earthquake feature samples. We evaluate the impact of major earthquake feature oversampling on prediction performance using various classification models, including AdaBoost, Random Forest, Support Vector Machine, and K-Nearest Neighbors. Results demonstrate that oversampling significantly enhances the model′s identification accuracy, recall, and R score, with an average R score above 0.6. This represents an improvement of about 0.2 compared to the current earthquake prediction levels in China. This demonstrates that oversampling large earthquake features can uncover hidden feature distribution patterns, contributing positively to earthquake prediction. Additionally, comparisons with Epidemic-Type Aftershock Sequence (ETAS) model reveal that the two have strong agreement in identifying high-risk areas for seismic activity. Both the machine learning methods and ETAS model indicate there is a high probability of M≥6 earthquakes occurring in the southern Sichuan-Yunnan region during the period from August 2024 to August 2025, highlighting the need for ongoing monitoring and research.