Intelligent earthquake prediction using animal vocal behavior analysis based on machine learning and deep learning...

Area of Science:

• Earthquake prediction
• Bioacoustics
• Machine Learning (ML)
• Deep Learning (DL)

Background:

• Traditional earthquake monitoring lacks timely warnings, leading to significant global losses.
• Early detection of seismic activity is crucial for mitigating environmental and infrastructural damage.
• Innovative approaches are needed to supplement existing seismic monitoring systems.

Purpose of the Study:

• To develop a novel framework for identifying earthquake precursors using animal bioacoustics.
• To leverage Machine Learning (ML) and Deep Learning (DL) models for analyzing animal vocalizations.
• To create a cost-effective and scalable early warning system for earthquakes.

Main Methods:

• Utilized historical animal audio recordings, enhanced with data augmentation.
• Extracted temporal and spectral features using the Librosa library.
• Implemented and compared ML algorithms (XGBoost, Random Forest, MLP) and DL models (RNN, LSTM, Bi-LSTM, GRU).

Main Results:

• Bidirectional LSTM (Bi-LSTM) achieved a test accuracy of 98.87% with an Area Under the Curve (AUC) close to 1.00.
• Deep learning models demonstrated superior performance in handling temporal dependencies in audio data.
• The model showed robustness to environmental noise and scalability with unseen datasets.

Conclusions:

• The proposed bioacoustics-based ML/DL framework effectively predicts earthquakes.
• This approach offers a cost-effective early warning system, especially for areas lacking advanced seismic infrastructure.
• Future work includes integrating IoT and edge computing for enhanced scalability.