Evaluating Medium Decision Tree Model, Support Vector Machine Rational Quadratic Gaussian Process Regression to Estimate the Total Organic Carbon of Shale Gas Reservoirs.

As a result of an energy crisis due to the Russian-Ukrainian war, the eyes of great countries such as America and others began to turn strongly towards exploiting unconventional resources to increase the oil and gas production. The first step in exploiting unconventional sources is to estimate the Total Organic Carbon (TOC). TOC measurements are expensive as well as time consuming, as samples of cuttings or core samples must be present to do the required lab tests. This issue encouraged the researchers to develop mathematical correlation to estimate the TOC. The paper aims at evaluating three of machine learning models namely medium decision tree model (MDT), support vector machine (SVM) and rational quadratic Gaussian process regression (GPR) learned based on well logs data for estimating the TOC. To reach this target, 334 datasets of TOC a function of gamma ray, formation resistivity and sonic transit time. The results showed that rational quadratic Gaussian process regression (GPR) has higher accuracy than other models in estimating TOC. GPR achieved correlation coefficient of 0.91 with root mean square error (RMSE) of 1.01% and mean average error (MAE) of 0.74%. [ABSTRACT FROM AUTHOR]