Prediction of felt intensity from ground motion parameters using artificial neural network method

Öztürk , Seda
Earthquakes are natural phenomena that cause ground shaking and deformations due to the nature of the Earth's surface, which is composed of tectonic plates. The sudden release of energy on these tectonic plates results in earthquakes. One of the ways to measure ground shakings is the macroseismic (or felt) intensity. There are various studies on the correlation between felt intensity and ground motion parameters. Most of them involve a linear regression method to find an empirical formula for this relation. However, assuming a linear correlation may not the best approach since the independent variables affecting intensity values show highly non-linear behaviour. Therefore, a more flexible model capturing the complexities of these independent variables should be constructed. In this thesis, initially, principal component analysis (PCA) is applied to identify main independent variables that affect felt intensity. Based on the results of PCA and expert knowledge, various artificial neural network (ANN) models are built. Feedforward backpropagation method is used with different combinations of input variables to study the best predictions of MMI. Most of the ANN models resulted in better MMI estimations than those provided in the literature.