Mechanical properties of rocks can significantly affect energy resource recovery and development. Uniaxial compressive strength (UCS) and Young’s modulus (E) are key parameters in rock mechanics for designing geotechnical structures. In this study, it is shown that a new practical machine learning tool known as random forest (RF) model can be used for variable importance measurements (VIMs) among various physical and mechanical properties of rocks. Moreover, RF is a predictive model which can estimate E and UCS based on selected variables by VIMs. Therefore, in this study first VIMs of RF were applied to assess various rock properties (porosity (n), point load index (Is(50)), P-wave velocity (Vp), and Schmidt hammer rebound number (Rn)) and to choose the best predictors to model E and UCS. Results of VIM, assisted by Pearson correlation, demonstrated that Vp is the most effective variable for the prediction of both E and UCS. The most effective variables (Vp-Rnfor E, and Vp-Is for UCS) were selected as inputs of RF model for the E and UCS predictions. Outputs in the testing stage of the models verified that RF can yield a satisfactory prediction of both E and UCS, with correlations of determination (R2) of 0.91 and 0.93, respectively. For comparison purposes, multivariable regression and generalized regression neural networks were used for E and UCS prediction. According to these results, developing a nonlinear inter-dependence approximation among parameters for variable selection and also a non-parametric predictive model by soft computing methods such as RF can potentially be further employed as a reliable and accurate technique for evaluation and estimation of complex relationships in rock mechanics.