A Comparison of Machine Learning Processes for Classification of Rock Units Using Well Log Data

Abstract

Summary This work aims to define a comparison between a Kohonen SOM, an euclidean and a mahalanobean classificators. This comparison uses two well log data from a synthetic syneclises sedimentary basin type. It is remarkable that the Mahalanobis classifier produced a higher error when compared to the Euclidean classifier and the SOM. The SOM presented better results for the two synthetic examples, with an error of 0.7% for the first well and 1.5% for the second. In contrast, Mahalanobis and Euclidean classifiers presented an error of 18.3% and 1.7% respectively for the first well and 11.3% and 6% for the second.