Identifying Potential Areas for Oil and Gas Well Location Planning Using Support Vector Machine Algorithm

To fulfill national energy needs for the National Energy Grand Strategy (GSEN), it is necessary to increase the productivity of oil and gas exploration by involving technology that provides alternative solutions, cuts work time, and overcomes the risk of failure. This research aims to identify potential areas for planning new oil and gas well locations using a machine learning algorithm called Support Vector Machine (SVM). This research chooses four splitting ratios of 80:20, 75:25, 60:40, and 50:50 on training and testing data to produce four models and to identify the most robust model for Blora Regency. The algorithm involves fourteen conditioning parameters comprising altitude, slope, aspect, distance from the river network, land cover, distance from the road network, soil type, Normalized Difference Vegetation Index (NDVI), clay mineral index, iron oxide index, surface temperature, complete Bouguer anomaly (CBL), distance from the fault, and rock type. This research uses the confusion matrix and the ROC-AUC to evaluate all models and determine the best one. The result witnesses the best model is SVM 75:25 with an accuracy (Acc), sensitivity (Sen), specificity (Spe), and predictive value (PPV) of 0.8333; Matthew's correlation coefficient and Cohen's kappa of 0.6667; and area under the curve (AUC) of 0.9444. In addition, the conditioning parameter contributing the most significant influence on the best model is the slope equal to 100%.