Formation Resistivity Prediction Using Decision Tree and Random Forest

Formation resistivity (R_t) is a vital property for formation evaluation and calculation of water saturation and hydrocarbon in places. R_t can be estimated using core analysis and well logging. However, these processes are expensive and time-consuming. In addition, due to tool failure, and poor wellbore conditions, part of the well logging records may be missed. Hence, the objective of this paper is to predict the true formation resistivity in complex carbonate sections using decision tree (DT) and Random Forests (RF) machine learning (ML) techniques as a function of available well logging data. A data set of 5500 data points were collected from two vertical wells in carbonate formation. The data includes gamma-ray, bulk density, neutron density, compressional wave transit time, shear transient time, and the corresponding R_t. Data from Well-1 were used to develop the DT and RF models with training to the testing splitting ratio of 70:30. Dataset from Well-2 was used to validate the optimized models. The results showed the capabilities of the ML models to predict the formation resistivity from well-logging data. The correlation coefficient (R) between the actual and the predicted output values and the root mean square error (RMSE) was used to evaluate the models performance. R value for the RF model was found to be 0.99, and 0.98 for the training and the testing stages with a validation R value of 0.94. The RMSE for the developed models was less than 0.38 for training, testing, and validation stages. Using ML to predict the formation resistivity can fill the missing gaps in log tracks and save money by removing resistivity logs running in all offset wells in the same field.