Deep Learning Approach in Seismology: Enhancing Earthquake Forecasting using K-Means Clustering and LSTM Networks

Abstract

Located in the subduction zone of four tectonic plates, the high occurrence of seismic events is a severe threat in Indonesia. Mitigating the adverse effects of such disasters is essential to forecast the likelihood of future earthquakes. Consequently, developing a robust method of forecasting future earthquakes is critical to facilitate prevention and mitigation efforts. A reliable earthquake prediction method is necessary to reduce the after-effects to the greatest extent possible. This study utilises historical seismic and proposes innovative data pre-processing methods using K-means clustering to build a Long Short-Term Memory (LSTM) model for earthquake forecasting to overcome high-disparity locations. Four LSTM layers are embedded with adjusted fine-tuned network hyperparameters to enhance forecasting accuracy. The results attain 0.379816, 0.616292, and 0.414586 for Mean Square Error (MSE), Root MSE, and Mean Absolute Error, respectively, providing significant insights into earthquake prediction. In addition, predicted seismic occurrences are plotted on a map to display their geographic location within the specified research region. This research provides significant value in facilitating the efficient distribution of resources, such as evacuating residents impacted by earthquakes or reinforcing buildings and infrastructure, for emergency responders and policymakers.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint