Spatial Machine Learning Algorithms to Discover Prospective Oil and Gas Wells Locations Based on Surface Driving Factors

Abstract

Indonesia faces challenges in transitioning its energy sector, aiming to shift from coal to natural gas, achieve net zero emissions with renewable energy, and overcome geographical complexity obstacles, diverse cultural perspectives, and a developing regulatory framework. To address these issues, the government actively studies the Grand National Energy Strategy in enhancing petroleum and fuel refineries. This research aims to expand the academic approach by utilizing spatial technology and machine learning to optimize the new oil and gas well placement determination and meet the high-demand resources. Four algorithms, support vector machine (SVM), random forest (RF), artificial neural network (ANN), and k-nearest neighbor (KNN), with four training and testing splitting scenarios (80:20, 75:25, 60:40, and 50:50) are used to produce probability map of the wells site suitability along with fourteen surface driving factors related to the environmental agreement. The outcome indicates that the 80:20 RF model demonstrated excellence, achieving a 0.95 accuracy, 1.00 sensitivity, 0.90 specificity and Cohen’s Kappa, 0.91 precision, and 0.99 area under the curve, showcasing the optimal fit with validation data. The four surface driving factors with the highest important index indicate that the well placement is sensitive to historical disaster, ease of accessibility, and hydrocarbon sourcing.