Predicting Photoelectric Logs in Challenging Conditions Using Machine Learning and Statistical Analysis

Abstract The photoelectric (PEF) log measures the photoelectric absorption factor, pivotal for determining rock matrix properties. High absorption factor values are typical in limestones, dolomites, clay, iron-bearing minerals, and heavy minerals, whereas sandstones exhibit lower values. In this study, actual photoelectric logs were gathered from the field alongside various other logs such as gallons per minute (GPM), standpipe pressure (SPP), rate of penetration (ROP), and bulk density (RHOB). Utilizing a suite of machine learning regression techniques—ridge regression, linear regression, support vector machines (SVM), polynomial regression, random forest, and decision tree—this research aimed to predict the photoelectric logs using porosity and other log data as inputs. The effectiveness of these models was confirmed through their strong predictive accuracy relative to actual log values. The ensemble of regression models demonstrated significant correlation coefficients and low root mean square errors, illustrating their robust capability to predict photoelectric data at various depths based on available drilling data.