Prediction of the Peak Ground Acceleration for Zagros Earthquakes Using ANFIS and Data Partitioning Approach

In this research, the prediction of Peak Ground Acceleration (PGA) is investigated through training the Adaptive Neuro Fuzzy Inference System (ANFIS) network using data partitioned into inlier and outlier groups. The partitioning procedure is based on an automated method which uses K Nearest Neighbor (KNN) search and Local Linear Model Tree (LOLIMOT) methods. The mentioned proved to enhance the prediction procedure at least 36 percent with respect to normal training mode with no data partitioning. Hereafter, a PGA catalogue reported by the Iranian ground acceleration network with 1571 records was used and the trained network was used for predicting the PGA map around the Mormori, 2014 earthquake epicenter. We used spatial information of the epicenter and the site, earthquake magnitude, Vs30, depth of the hypocenters and the epicentral distance foe the sites. The resulted map adapts well to the official report of the mentioned earthquake for the PGA analysis published by the International Institute of Earthquake Engineering and Seismology (IIEES). Finally, it is concluded that using machine learning algorithms could be beneficial in the cases where adequate data-sets are not provided by the seismological networks specifically in the concept of the strong ground motion analysis.