<![CDATA[Accurate prediction of shear wave velocity (Vs) is a crucial element in seismic modeling, amplitude variation with offset (AVO) analysis, and comprehensive reservoir characterization. However, the availability of direct measurement data from sonic logs is frequently limited, particularly in older exploration wells or those with incomplete datasets. This study presents an in-depth comparative analysis between machine learning (ML) algorithms and conventional deterministic methods for estimating Vs values in an exploration well. Three ML algorithms, namely XGBoost, K-Nearest Neighbor (KNN), and Support Vector Machine (SVM), were evaluated against actual measured data and compared with both the standard Castagna equation and a locally calibrated version optimized through sensitivity analysis. Quantitative results demonstrate that the XGBoost algorithm delivers the highest predictive performance, achieving a Coefficient of Determination (R2) of 0.991, alongside the lowest error metrics (MAE = 62.3 m/s; RMSE = 92.5 m/s). Conversely, the standard Castagna equation exhibits significant inaccuracies in high-velocity zones exceeding 2500 m/s. Through sensitivity analysis, this study successfully optimized the empirical constants, producing a calibrated Castagna equation. Although ML models are statistically superior in capturing complex petrophysical patterns, the modified Castagna equation proves to be a robust and highly practical alternative for exploration scenarios with constrained datasets. This study recommends deploying XGBoost for field-scale developments to simultaneously process massive multi-well data without requiring time-consuming individual recalibrations. Ultimately, integrating ML techniques and local coefficient optimization significantly minimizes uncertainties in hydrocarbon exploration through precise elastic rock parameter estimation.]]>
