Automatic Parameter Selection Using K-Fold Test: Application to Interpolation Algorithm
Nihed El Allouche, Kemal Özdemir, Lee West, Ben Veitch · 2020
Summary We propose to use the -fold cross-validation test from statistics and machine learning to automate the parameter selection for a matching pursuit Fourier interpolation algorithm. Matching pursuit-based interpolators are widely used to regularize nonuniform seismic data and interpolate data with large gaps. Automatic parameter selection helps reduce the testing time and provides an objective metric to evaluate the results. The proposed method of automation splits the input traces into random subsets where each subset has the same number of traces and devises tests such that, at each test, one subset is kept for validation and the remaining subsets are used for training. Averaged interpolation error computed over all training sets gives an error metric for choosing the optimum interpolation parameter objectively. As an example, we use the k-fold test to estimate automatically the optimum number of iterations required to interpolate an irregularly sampled data set.