Earthquake prediction has been a challenging research area. In this work, we plan to compute sixty seismic features by employing seismological concepts, such as Gutenberg-Richter law, seismic rate changes, foreshock frequency, seismic energy release, total recurrence time. Further, Maximum Relevance and Minimum Redundancy (mRMR) criteria can be applied to extract the relevant features. Using these feaures, a Support Vector Regressor (SVR) and Hybrid Neural Network (HNN) based classification system could prove to be optimal for predicting earthquakes. HNN is a step wise combination of three different Neural Networks, supported by Enhanced Particle Swarm Optimization (EPSO), to offer weight optimization at each layer.
In this study, three different regions, namely Hindukush, Chile, Southern California have been selected for prediction of earthquakes of magnitude 5.0 and above. The same regions selected in the precedent studies are also considered for this project. The advantage of selecting the same regions is that results can be compared in the end, so as to prove the superiority of suggested methodology.
Earthquake catalogs of these regions have been obtained from United States Geological Survey (USGS) for the period from January 1980 to December 2016.
In this project, features are also referred as seismic parameters. These seismic parameters are calculated mathematically and are based upon well-known geophysical and seismological facts. Some of the features are listed below as an example.
- Seismic energy release
- Time of n events
- Seismic rate changes
- Probability of earthquake occurrence
- Deviation from Gutenberg-Richer law
Many other parametric and non parametric features have been used to build the earthquake prediction model.