Earthquake Classification Using Resnet 50 Model: A Machine Learning Approach

ABSTRACT: Earthquakes can result in significant loss and damage to life and infrastructure, which requires rapid and accurate detection. Traditional methods often face challenges in classifying earthquake signals from noise especially in seismically active zones. This study explores the application of the ResNet-50 deep-learning convolutional neural network model to classify earthquake signals using seismic waveforms. The model is trained on the Italian earthquake dataset containing 1.2 million three-component traces from approximately 50,000 earthquakes and 130,000 noise traces recorded between 2005 and 2020. ResNet-50 enhances feature representation better gradient flow during backpropagation through residual connections. The Binary Cross Entropy (BCE) loss function and ADAM optimizer is used in this study with spectral analysis aiding to extract frequency features. The model achieves an accuracy ranging from 0.79 to 0.95, with precision and recall between 0.79 to 0.97, demonstrating its effectiveness in accurately identifying seismic events amid background noise. These findings highlight the potential of the ResNet-50 model to enhance earthquake monitoring and early detection systems. 1. INTRODUCTION One of the most impressive geological phenomena, earthquakes, can have disastrous consequences for life and infrastructure. Earthquakes continue to be among the most erratic natural disasters. A summary of the 20 years’ worth of catastrophes shows 552 earthquakes or 8% of all disasters globally. These earthquakes rank third after storms (2043 events, or 28% of the total) and floods (3254 events, or 44%) (Mavrouli et al., 2023). In addition to intense ground motion and seismic events, secondary effects, primarily landslides and tsunamis, are also blamed for related calamities causing loss of life and damage to infrastructures. Accurate classification of earthquake signals is important in applications of earthquake early warning, monitoring and seismic data processing. This work aims to train a convolutional neural network (CNN) with seismic data to enable it to identify signals as noise or earthquake. Recent developments in Machine learning have led to complex neural network architectures. These architectures perform exceptionally effectively in pattern recognition tasks. Deep learning algorithms like CNN effectively segment and classify tasks, including image recognition, signal processing, and medical imaging. These algorithms can learn complex patterns and features from the data, making them a great tool to apply to waveform analysis for robust end, efficient earthquake signal detection tasks.