Seismic detection using submarine cable polarization signals with machine learning

This repository contains the code and datasets used in the paper “Seismic detection using submarine cable polarization signals with machine learning”. It provides a complete workflow for processing submarine cable polarization time-series data, extracting features, performing statistical analyses, machine learning classification, and deep learning with LSTM autoencoders. The repository is organized into several main folders: preprocessing/ – scripts to clean and segment raw seismic and SOP data. benchmark_analysis/ – implementations of benchmark methods for time-series analysis, including PSD and STA/LTA. machine_learning/ – feature extraction and machine learning classification of earthquake vs non-earthquake events. autoencoder/ – training and testing of LSTM autoencoders on time-series data. earthquake_properties/ – analysis of earthquake event properties, including travel-time comparisons and mapping. utilities/ – core utility functions and custom classes used throughout the pipeline. All datasets required to reproduce the analyses are provided in .pkl format, including feature dataframes and segmented time-series lists. The data are organized to ensure reproducibility of statistical, machine learning, and deep learning experiments.