Earthquake Prediction Learning Using the Least Alarm Method

An approach to organization of a system for automatic earthquake prediction is developed. This approach is not bound to any particular type of data, and all data on seismotectonic processes are displayed using the spatial and space-time fields. The prediction field is learned from the feature fields and the sample of predicted earthquakes. The learning method proposed here is the least alarm method. The results of modeling of earthquake prediction for the regions of Japan and Mediterranean are presented. Modeling implied the analysis of prediction fields reflecting stationary and dynamic properties of the seismic process. The results of modeling demonstrate the efficiency of application of the least alarm method to earthquake prediction.