A comparison of binary and multiclass support vector machine models for volcanic lithology estimation using geophysical log data from Liaohe Basin, China

Lithology estimation of rocks, especially volcanic lithology, is one of the major goals of geophysical exploration. In this paper, we propose the use of binary and multiclass support vector machine models with geophysical log data to estimate the volcanic lithology of the Liaohe Basin, China. Using neutron (CNL), density (DEN), acoustic (AC), deep lateral resistivity (RLLD), and gamma-ray (GR) log data from 40 wells (a total of 1200 log data points) in the Liaohe Basin, China, we first construct the binary support vector machine model to classify volcanic rock and non-volcanic rock. Then, we expand the binary model to a multiclass model using the approach of directed acyclic graphs, and construct multiclass models to classify six types of volcanic rocks: basalt, non-compacted basalt, trachyte, non-compacted trachyte, gabbro and diabase. To assess the accuracy of these two models, we compare their predictions with core data from four wells (at 800 different depth points in total). Results indicate that the accuracy of the binary and multiclass models are 98.4% and 87%, respectively, demonstrating that binary and multiclass support vector machine models are effective methods for classifying volcanic lithology.