I study the predictability of large events in self-organizing systems. I focus on a set of models which have been studied as analogs of earthquake faults and fault systems, and apply my methods based on 2 class pattern classification. In my methodology, first class contains seismics with magnitude smaller than M_thr and second class contains seismics with magnitude larger than M_thr. First I extract features of samples from foreshock zone for each class, and then reduce the dimensionality of extracted features. For classification of these samples based on their reduced features, I use MLP neural network. For forecasting purpose, trained MLP network must classify extracted foreshock features of future events. In this way my method could be used for forecasting of majority of magnitude of next events at exactly certain time (several months) before occurrence of that event, unfortunately the location of epicentres of events must known. As I know the PI method can be used for long-term forecasting of next earthquakes in PI map, so first I locate future earthquakes with PI method and then use my method for accurate forecasting of the time and locations of next earthquakes at PI-alarmed locations. I validated my method with ROC diagram of MLP classifier for testing samples. Results show that my method can be used for short-term forecasting of large earthquakes.

Bookmark: