PREDICTING WELL LOGS USING MACHINE LEARNING TECHNICS

Abstract

Accurate and continuous well log data are critical for subsurface characterization and reservoir evaluation in the petroleum industry. However, missing or degraded Gamma Ray (GR) logs—essential for lithological identification and shale content estimation—pose significant challenges to geoscientists. This study proposes a machine learning-based framework for predicting GR log values using other routinely acquired well logs, including bulk density (RHOB), neutron porosity (NPHI), and compressional slowness (DT), with a focus on data from the Hassi Terfa field in southeastern Algeria. Several machine learning algorithms were implemented, including Linear Regression (LR), Random Forest (RF), Gradient Boosting (GB), Artificial Neural Networks (ANN), Support Vector Regression (SVR), and K-Nearest Neighbors (KNN). Model performance was assessed using metrics such as R2, RMSE, and MAE. The findings show that the ensemble methods, especially the Random Forest algorithm, performed well, as demonstrated by its R2 reaching 0.94, RMSE of 0.031, and MAE of 0.0009. This highlights its ability to effectively handle complex, nonlinear relationships in petrophysical data. This work highlights the potential of machine learning to enhance subsurface data quality, reduce reliance on costly re-logging operations, and support more reliable geological interpretations. It also sets the stage for further research into advanced modeling techniques and broader geological applications.