Applicability of Statistical Learning Algorithms for Seismic Attenuation Prediction

ABSTRACT: Frequent occurrence of seismic activities has drawn attention of various researchers, cutting across science and engineering discipline. The domain of earthquake geotechnical engineering has extended from determination of attenuation of seismic wave to providing suitable foundation under liquefied and unliquefied soil conditions. The seismic data can be fully intrepreated by use of seismic attenuation through the rock. However, the relation between seismic attenuation and rock properties is very complex. The physical properties like porosity, permeability, grain size and clay content affects the seismic attenuation. Various statistical effeorts have been made to predict the seismic attenuation based on the above physical properties. In the context, this paper describes a statistical learning algorithms applied for seismic attenuation prediction based on physical properties of rock. The support vector machine (SVM) that is firmly based on the theory of statistical learning theory has been applied with porosity, permeability, grain size and clay content as the inputs. In this study, Gaussian, polynomial and spline functions have been used as kernels for SVM algorithms. The results have been compared with that obtained using artificial neural network (ANN). The predicted results show the ability of learning algorithms to build accurate models for the prediction of seismic attenuation with strong predictive capabilities. A comparison between the SVM and ANN models demonstrates that the SVM model is superior to ANN model in predicting seismic attenuation.