Machine Learning-Based Classification of Seismic Events:A Case Study of seismic events in Jilin Province, NE China

Fangyu Ren¹Hao Liang^2*Hongyan Zhang¹Tingting Wang³Qingfeng Ruan¹Fan Zhang¹Yu Wang⁴

• ¹Jilin Earthquake Agency, Changchun, China
• ²China Earthquake Networks Center, Beijing, China
• ³Institute of Geophysics China Earthquake Administration, Beijing, China
• ⁴Shandong Earthquake Administration, Jinan, China

The ability to rapidly distinguish between natural and non-natural seismic events is of great significance for the construction of seismic catalogues and seismic hazard assessments. In this study, we examine a collection of 898 events in both time and frequency domains with local magnitudes of 1.5≤ML≤3.5 from the Jilin Seismological Network's seismic event catalogue between 2013 and 2024. An 87-dimensional feature set was constructed, including spectrum amplitude, maximum P/S amplitude ratio, high/low-frequency energy ratio, inflection point frequency, waveform duration, complexity, zero-crossing rate, and instantaneous frequency. Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), and Backpropagation Neural Network (BPNN) algorithms were employed to classify artificial blasts, mining collapses, and tectonic earthquakes. The analysis results show that all three methods achieve classification accuracies exceeding 94% for multi-class events. The accuracy rate reached 84% in cross-regional validation scenarios. Comparative analysis shows that SVM performs better than other models. All extracted features contributed to classification efficacy, with P-/S-wave spectrum amplitude and maximum P/S amplitude ratio exhibiting the highest feature importance. This work demonstrates that machine learning can enable the rapid and accurate identification of multiple types of moderate-to-small magnitude seismic events.