Enhancing the classification of seismic events with supervised machine learning and feature importance

Area of Science:

• Seismology
• Geophysics
• Machine Learning Applications

Background:

• Accurate classification of seismic events is vital for geological understanding, hazard mitigation, and public safety.
• Differentiating natural earthquakes (EQ) from man-made quarry blasts (QB) presents a significant challenge in seismology.

Purpose of the Study:

• To propose and evaluate a machine learning (ML) approach for discriminating between seismic events (EQs and QBs).
• To identify the most effective features for accurate seismic event classification using feature selection and importance analysis.

Main Methods:

• Integration of diverse seismic features into a unified dataset.
• Training and benchmarking of linear and nonlinear supervised ML models.
• Application of feature selection techniques to pinpoint critical discriminating features.

Main Results:

• A machine learning model achieved 100% discrimination accuracy between EQs and QBs.
• Feature importance analysis identified corner frequency, event power, and spectral ratio as the most crucial features.
• The developed model outperformed several benchmark linear and non-linear models.

Conclusions:

• Machine learning, particularly with carefully selected features, offers a highly effective method for distinguishing seismic events.
• The identified key features provide valuable insights into the physical differences between earthquakes and quarry blasts.
• This approach enhances seismic event classification accuracy and reliability.