A Step Forward to Automatic Seismic Focal Classification Using AI: Meta-Heuristic Wrapper and Shallow Learning Applications

In this paper, a new step forward to the application of advanced artificial intelligence techniques in order to predict an automatic seismic focal mechanism classification from strong ground motion records, in a more efficient way, is presented. In order to effectively reduce computing effort and to minimize the feature-space characteristics extracted from seismic data, this work proposes a multiclass classification method in a comprehensive automatic framework, using selected metaheuristic-based wrapper strategies and some selected shallow learning classifiers, permitting to predict the seismic focal mechanism that produces an earthquake, in a very efficient way, with a considerable reduction of the feature-space characteristics that need to be analysed. The feature-space reduction is achieved applying nature-inspired algorithms such as a genetic evolutionary and a swarm intelligence algorithm, combined with seven nearest neighbours’ machine learning classifiers. The best scheme obtained very successful mean AUC scores (0.807 and 0.940) for the training and test stages, meaning that it is possible to consider this proposed technique as a very powerful tool for predicting seismic focal mechanisms of a recorder earthquake, almost in real time, with interesting future applications in early warning systems or other similar seismic applications.