GeoAI Seismic Event Classification

Overview

This project develops a machine learning pipeline for classifying seismic waveforms as either earthquake events or noise using the STEAD (Stanford Earthquake Dataset).

The workflow includes:

• Reading seismic waveforms from HDF5 files
• Feature extraction from 3-component seismic traces
• Dataset generation
• Machine learning classification
• Model evaluation and interpretation

---

Dataset

STEAD Dataset

Classes Used:

• Noise Traces (Chunk 1)
• Earthquake Traces (Chunk 4)

Dataset Size Used:

• 5000 Noise Traces
• 5000 Earthquake Traces

Total: 10,000 Traces

---

Features Extracted

• Mean
• Standard Deviation
• Maximum Amplitude
• Minimum Amplitude
• Peak-to-Peak Amplitude
• RMS Amplitude
• Signal Energy
• Zero Crossing Rate
• Skewness
• Kurtosis
• Dominant Frequency
• Spectral Centroid

---

Machine Learning Model

Random Forest Classifier

Training/Test Split:

• 80% Training
• 20% Testing

---

Results

Accuracy: 98.8%

Evaluation Metrics:

• Precision
• Recall
• F1 Score
• Confusion Matrix

---

Generated Outputs

• rf_stead.pkl
• confusion_matrix.png
• feature_importance.png

---

Technologies Used

• Python
• NumPy
• Pandas
• SciPy
• Scikit-Learn
• ObsPy
• H5Py
• Matplotlib

---

Future Work

• XGBoost Benchmarking
• Deep Learning (1D CNN)
• Real-Time Seismic Event Detection
• Spectrogram-Based Classification