High resolution estimation shear wave velocity and rock elastic properties using integration of the individual and hybrid machine learning models

• A total of fifteen machine learning algorithms were employed for shear wave velocity and elastic rock properties estimation • A novel Weighted Averaging Committing Machine (WACM) was formulated for shear wave velocity estimation, surpassing all individual models • The study underscores the effectiveness of ensemble learning in estimating geophysical properties, highlighting the significance of integrating diverse model perspectives to enhance reliability. • The predicted Vs values were used to compute dynamic elastic moduli, which were then transformed into static moduli through empirical correlations, providing a reliable methodology for static parameter estimation. • The proposed framework demonstrates exceptional predictive accuracy and resilience, with substantial potential for practical application in geomechanics, petroleum engineering, and seismic analysis In this research, a robust machine learning-based framework was established to estimate shear wave velocity (Vs) and elastic rock properties through both individual and ensemble predictive models. The data utilized in this investigation were sourced from two gas wells located in southern Iran, encompassing a total of fifteen machine learning algorithms, which included tree-based, kernel-based, neural network, linear, and distance-based models. Among the individual models, the Random Forest algorithm attained the highest Coefficient of Determination (R²) 0.94. In contrast, the General Regression Neural Network (GRNN) and Radial Basis Function (RBF) models exhibited comparatively lower accuracy, yielding R² values of 0.8478 and 0.848, respectively. To enhance the predictive precision, a Weighted Averaging Committing Machine (WACM) was formulated, employing Ridge Regression as a meta-learner within a stacking ensemble framework. This hybrid approach significantly surpassed all individual models, achieving a notable test R² of 0.97. Utilizing the predicted Vs values, we computed dynamic elastic moduli specifically Young's modulus, Poisson's ratio, Bulk modulus, and Shear modulus using standard petrophysical equations based on density and sonic logs. These dynamic properties were subsequently transformed into static moduli through empirical correlations established by Ranjbar-Karami et al. 3 for carbonate reservoirs, providing a reliable methodology for static parameter estimation. The proposed framework demonstrates exceptional predictive accuracy, resilience across various model types, and substantial potential for practical application in geomechanics, petroleum engineering, and seismic analysis. The findings underscore the efficacy of ensemble learning in estimating geophysical properties and highlight the significance of integrating diverse model perspectives to enhance reliability.