Supervised learning to detect salt body

Summary In this paper we are presenting a novel workflow to detect salt body base on seismic attributes and supervised learning. The machine learning algorithm Extremely Random Trees Ensemble is used to train and automatically identifying and classify salt regions. We have used a complex synthetic seismic dataset from phase I model of the SEG Advanced Modeling Corporation (SEAM), that represents deepwater regions of Gulf of Mexico. This dataset has very low frequency and there are sediments locations with similar amplitude value than salt body. After a first step of our proposal, where machine learning is applied directly to the seismic data, we obtained accuracy values of around 80%. A second (post-processing) step brings up accuracy to around 95%. We conclude that machine learning is a promise mechanism to identify salt bodies on seismic data when the selected model exhibits enough capacity to model the complex decision boundaries needed during class discrimination.