Applicability of machine learning for shear-wave velocity prediction from conventional well logs: the LSBoost approach.

Saved in:

Bibliographic Details
Title:	Applicability of machine learning for shear-wave velocity prediction from conventional well logs: the LSBoost approach.
Authors:	WIBOWO, Rahmat Catur¹ rahmat.caturwibowo@eng.unila.ac.id, ARBY, Fadsyah Muhammad², MULYATNO, Bagus Sapto², DEWANTO, Ordas², KUMALASARI, Isti Nur², SARKOWI, Muh²
Source:	Turkish Journal of Earth Sciences. 2026, Vol. 35 Issue 3, p236-249. 18p.
Subjects:	Boosting algorithms, Geophysical well logging, Geotechnical engineering, Machine learning, Multiple regression analysis, Model validation, Support vector machines
Abstract:	Shear-wave velocity (Vs) is one of the most critical parameters for determining geomechanical properties and basin overpressure. However, assessing Vs via techniques like core analysis requires considerable effort and expense. This study predicts Vs using several approaches and compares the accuracy levels of all models. For this objective, the multiple linear regression, multiple linear stepwise regression, support vector machine, and least-squares boost (LSBoost) methodologies were selected. The six well-logging data inputs of density (RHOB), gamma-ray (GR), deep resistivity (ILD), acoustic wave velocity (Vp), shale volume (VCL), and water saturation (SW) were selected as effective variables, whereas Vs was regarded as the output. The model was developed using data from the RCW-1 well and evaluated through 5-fold cross-validation and independent blind cross-well validation. The LSBoost model demonstrated robust and stable in-well performance, achieving a mean coefficient of determination (R² ) of 0.958 ± 0.003 and root mean square error (RMSE) of 56.48 ± 1.58 m/s, indicating effective learning without evidence of overfitting. Feature importance and Spearman rank correlation analyses consistently identified ILD and Vp as the most influential predictors, confirming the presence of nonlinear monotonic relationships between the input logs and Vs. However, blind application of the RCW-1-trained model to the independent RCW-2 well resulted in reduced predictive accuracy (R² = 0.266, RMSE = 287.95 m/s), reflecting geological and petrophysical domain shifts between wells rather than model inadequacy. Comparisons with empirical Castagna and Greenberg–Castagna correlations showed that while the empirical models provided relatively stable baseline predictions across wells, LSBoost significantly outperformed them within a consistent geological domain. These results highlight the potential of LSBoost for accurate Vs prediction when representative training data are available and emphasize the importance of multiwell datasets to improve cross-well generalization. [ABSTRACT FROM AUTHOR]
	Copyright of Turkish Journal of Earth Sciences is the property of Scientific and Technical Research Council of Turkey and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Shear-wave velocity (Vs) is one of the most critical parameters for determining geomechanical properties and basin overpressure. However, assessing Vs via techniques like core analysis requires considerable effort and expense. This study predicts Vs using several approaches and compares the accuracy levels of all models. For this objective, the multiple linear regression, multiple linear stepwise regression, support vector machine, and least-squares boost (LSBoost) methodologies were selected. The six well-logging data inputs of density (RHOB), gamma-ray (GR), deep resistivity (ILD), acoustic wave velocity (Vp), shale volume (VCL), and water saturation (SW) were selected as effective variables, whereas Vs was regarded as the output. The model was developed using data from the RCW-1 well and evaluated through 5-fold cross-validation and independent blind cross-well validation. The LSBoost model demonstrated robust and stable in-well performance, achieving a mean coefficient of determination (R² ) of 0.958 ± 0.003 and root mean square error (RMSE) of 56.48 ± 1.58 m/s, indicating effective learning without evidence of overfitting. Feature importance and Spearman rank correlation analyses consistently identified ILD and Vp as the most influential predictors, confirming the presence of nonlinear monotonic relationships between the input logs and Vs. However, blind application of the RCW-1-trained model to the independent RCW-2 well resulted in reduced predictive accuracy (R² = 0.266, RMSE = 287.95 m/s), reflecting geological and petrophysical domain shifts between wells rather than model inadequacy. Comparisons with empirical Castagna and Greenberg–Castagna correlations showed that while the empirical models provided relatively stable baseline predictions across wells, LSBoost significantly outperformed them within a consistent geological domain. These results highlight the potential of LSBoost for accurate Vs prediction when representative training data are available and emphasize the importance of multiwell datasets to improve cross-well generalization. [ABSTRACT FROM AUTHOR]
ISSN:	13000985
DOI:	10.55730/1300-0985.2017