Facies classification is the process to determine the local rocks lithology by analyzing indirect measurements, such as well logs. Usually it is done manually by an interpreter. In this work, I am presenting an automatic method for facies classification by the use of feature engineering and gradient boosting trees. I used a set of classified well logs to train a multiclass machine learning model, and compared the predictions with both raw and processed features in a blind well. I could demonstrate that preparing the, by creating new features from the original well logs, such as their gradients, polar coordinates transformations, and clustering analysis, increased the predictions accuracy from 47% to 60%.
View full article as PDF (1.90 Mb)