Quake Prediction on Etna: A Multi-Parametric Deep Learning Approach

This thesis presents a data-driven framework for short-term earthquake prediction at Mount Etna, in collaboration with INGV, leveraging high-resolution geophysical datasets and advanced machine learning techniques, including deep learning.

Objective

Investigate the potential of machine learning methods to analyze seismic and related geophysical signals, focusing on predicting local earthquakes during 2018, a period of intense volcanic activity.

Tools & Technologies

• Python, Jupyter Notebooks
• Deep Learning (LSTM networks)
• Statistical modeling & Random Forest
• Time-series data preprocessing and feature engineering
• Data visualization and reporting

Methodology

1. Dataset and Preprocessing: Collected seismic, geophysical, and gas emission data; performed feature engineering and data cleaning.
2. Baseline Model: Random Forest for initial predictive performance.
3. Sequential Model: LSTM networks capturing temporal dependencies and nonlinear patterns.
4. Extended Architecture: Multi-LSTM models for enhanced performance.
5. Evaluation: Comparative analysis of model predictions to assess accuracy and robustness.

Key Takeaway

Demonstrates practical experience in end-to-end data projects, combining advanced machine learning, deep learning, and geophysical domain knowledge to support short-term earthquake forecasting at an active volcano.

Contents

• Thesis PDF: full document
• short presentation of goal and solution