Intelligent Classification of Low-Magnitude Earthquake Events: A Machine Learning Solution to Improve Seismic Monitoring Accuracy

This article introduces an open-source machine learning project aimed at solving the problem of accurately classifying low-magnitude earthquake events (M1-4). Through classic machine learning models and imbalanced data processing techniques, the project improves the quality of seismic catalogs, provides reliable data support for disaster assessment, and helps understand fault activity and seismic risk.

Low-magnitude earthquake identification faces multiple challenges: weak signals are easily overwhelmed by background noise; waveforms of different types of earthquakes (tectonic, collapse, blasting) vary greatly, and manual identification relies on experience; massive monitoring data is difficult to analyze manually in real time; there is a serious class imbalance problem in the data, which traditional methods struggle to handle effectively.

The project selects three classic models: logistic regression, random forest, and support vector machine (more robust and interpretable in scenarios with limited and imbalanced data); for the class imbalance problem, special strategies are adopted to ensure the recognition accuracy of all types of events and avoid the model favoring the majority class.

The data comes from the USGS public earthquake database; input features include magnitude, focal depth, geographic coordinates, etc.; the classification objective is to distinguish different types of low-magnitude earthquake events, improve the quality of seismic catalogs, and support seismic risk assessment.

Monitoring institutions can supplement existing systems to improve the efficiency and accuracy of low-magnitude event processing; researchers can use it as a starting point to explore complex models or applications in specific regions; a graphical interface is provided, allowing users without programming background to load CSV data, select models, and obtain classification results.

Follows best practices in machine learning engineering: data loading supports CSV format and validation; model switching function allows selection of appropriate algorithms; results are displayed in tables and support export; feature importance analysis helps understand key classification factors and provides references for seismic mechanism research.

Limitations: Classification accuracy depends on the quality of input data; not using waveform data limits the fineness; improvement directions: introduce waveform features, explore deep learning applications, and customize training models for specific geological regions.

Low-magnitude earthquake classification is an important but easily overlooked issue in the field of seismic monitoring. This project provides a practical solution through classic methods and imbalance processing, improving catalog accuracy and assessment reliability, and is worth trying for seismology researchers, monitoring institutions, and developers.