Stochastic Configuration Networks for Data Augmented Seismic Magnitude Prediction

Geophysical exploration yields a vast amount of monitoring data that accurately captures the geological dynamics of areas under investigation, providing a rich dataset for the analysis of seismic events. Despite a growing interest in utilizing monitoring data for event-centric information learning, challenges such as data imbalance significantly pose formidable obstacle to accurate information learning. In this study, a novel data augmentation method termed Adaptive Feature Expansion (AFE) is proposed to mitigate data imbalance. Additionally, an innovative randomized learning paradigm with incremental model configuration mechanism, Stochastic Configuration Networks (SCNs), is employed to learn seismic information and predict magnitude ranges. The experimental results indicate that AFE can effectively balance seismic data in the feature space and present excellent scalability. SCNs outperform traditional machine learning algorithms in terms of predictive performance. This work makes insightful contributions to geological hazard assessment.