Introducing stochastic recurrence interval to classification algorithms for identifying asperity patterns

Abstract

The introduction of stochastic earthquake recurrence times in feature vector is attempted for the identification of asperities in the area of Hokkaido, Japan, using machine learning algorithms. Seismicity attributes, feature selection algorithms, and class balancing techniques were used. The stochastic attributes of earthquake density in space, b-value, and the earthquake recurrence intervals were set as asperity identifiers. The study area was divided into 422 subareas, and in each one of them the aforementioned attributes were estimated for each subarea. For increasing the method efficiency a feature selection algorithm was utilized to indicate which of the selected attributes have the potential to contribute to the identification of asperities. A feature vector is presented, combining the attributes mentioned above, and well-known machine learning algorithms were used to identify the asperities locations. The performance of the method was tested with the 10-fold cross-validation technique and was found sufficient in means of F1 score.