The Implementation of Machine Learning in Lithofacies Classification using Multi Well Logs Data

Abstract

Lithofacies classification is a process to identify rock lithology by indirect measurements. Usually, the classification is processed manually by an experienced geoscientist. This research presents an automated lithofacies classification using a machine learning method to increase computational power in shortening the lithofacies classification process's time consumption. The support vector machine (SVM) algorithm has been applied successfully to the Damar field, Indonesia. The machine learning input is various well-log data sets, e.g., gamma-ray, density, resistivity, neutron porosity, and effective porosity. Machine learning can classify seven lithofacies and depositional environments, including channel, bar sand, beach sand, carbonate, volcanic, and shale. The classification accuracy in the verification phase with trained lithofacies class data reached more than 90%, while the accuracy in the validation phase with beyond trained data reached 65%. The classified lithofacies then can be used as the input for describing lateral and vertical rock distribution patterns.