Data Mining Techniques for Complex Formation Evaluation in Petroleum Exploration and Production: A Comparison of Feature Selection and Classification Methods

Abstract

Data mining techniques, especially classification methods, are receiving increasing attention from researchers and practitioners in the domain of petroleum exploration and production (E&P) in China. To extensively investigate the effects of feature selection and learning algorithms on the hydrocarbon reservoir prediction performance, taking three real-world multiclass problems as examples, namely formation evaluation of water-flooding interval, low resistivity reservoir, and gas zone from Chinese oil fields, this paper presents a comprehensive comparative study of both five feature selection methods including expert judgment, CFS, LVF, Relief-F, and SVM-RFE, and fourteen algorithms from five distinct kinds of classification methods including decision tree, artificial neural network, support vector machines(SVM), Bayesian network and ensemble learning. The results show that Relief-F and SVM-RFE can improve prediction performances more effectively than other methods, as well as C-SVC is the best classifier with generalization accuracy of 79.48%, 80.99%, and 75.01% respectively. Our studies suggest that the choice of classification methods should be more important than that of feature selection algorithms and the combination of SVMs and feature ranking should be preferred to other approaches for the complex reservoir evaluation using well logs.