Random-Forest Prediction of Shear-Wave Slowness from Conventional Well Logs for Geomechanical Applications: A Two-Well Case Study

Abstract

The aim of this study is to evaluate whether a random-forest model trained on conventional well logs can reconstruct missing shear-wave slowness (DTS) with sufficient transparency for downstream geomechanical calculations. Field-acquired DTS and 12 conventional curves from Well A were used for model development and progressive feature screening; an 11-variable subset was retained as a parsimonious input set, and Well Y was used as an external cross-block test. A re-audit of the prediction pairs embedded in the manuscript gave R2 = 0.97, MAE = 3.404 μs/ft, and RMSE = 5.718 μs/ft for the plotted Well A profile (n = 1873), and R2 = 0.91, MAE = 3.103 μs/ft, and RMSE = 3.658 μs/ft for Well Y (n = 255). Because the archived dataset lacks the raw feature-by-depth matrix, complete training settings, aligned density data, and core-based static-property measurements, depth-blocked retraining and numerical validation of mechanical properties could not be performed. The contribution is therefore an error-audited two-well DTS-completion case study and a transparent interface to potential geomechanical applications, not a validated hydraulic-fracturing design method.