Seismic facies classification with parametric and nonparametric statistics

Brian H. Russell

In this study, I compare parametric and nonparametric statistical methods for seismic facies classification. I will use an example that involves two seismic attributes, two facies and ten points. These seismic attributes are the normalized Vp/Vs ratio and the acoustic impedance, Ip. This problem includes several outliers and therefore illustrates the advantages and disadvantages of each method discussed. The classification techniques covered in this study consist of linear least-squares, k-nearest-neighbours (kNN), Quadratic Discrimination Analysis (QDA), Kernel Density Estimation (KDE) and the Deep Feedforward Neural Network (DFNN).