Full text loading...
-
Robust Evaluation of Fault Prediction Results: Machine Learning Using Synthetic Seismic
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, First EAGE Digitalization Conference and Exhibition, Nov 2020, Volume 2020, p.1 - 5
Abstract
Metrics to assess machine learning methods are necessary for evaluation of results and for comparisons to existing methods. For the segmentation of faults in seismic data, we suggest the use of a robust Jaccard metric that allows for small lateral inaccuracies in fault positioning. This error tolerance is necessary because interpretations are often inaccurate or subjective as a result of low seismic resolution, noise or other image deficiencies. The metric is used to evaluate results during the development of a 3D convolution neural network. In practice, this is done by applying new versions of the convolutional neural network to field data and by using metrics to compare machine learning results to the manual interpretations.