Elastic modulus of intact rock (E) plays a critical role in designing of civil, mining and rock engineering projects. Generally, this parameter is directly measured in laboratory in which complicated test tools and core specimens with high quality and fitting dimensions are needed. As an alternative, back propagation neural network (BPNN), radial basis function neural network (RBFNN) and multiple linear regression (MLR) techniques are used to estimate E in this study. For this purpose, measured data from the Azad and Bakhtiary dam sites in Iran are considered for models construction and evaluation. On the basis of minimum error index, neural networks with architecture 8-5-7-1 and 8-12-1 are found as the optimum networks for BPNN and RBFNN models, respectively. Comparing the performance of developed models is conducted using the five statistical indices. Accordingly, it is found that the performance of RBFNN model is somewhat better than BPNN model and both more realistic than the MLR mode. Also, the results of RBFNN and BPNN models gave higher conformity with the actual dada compared to the MLR model. Finally, parametric study shows that inputs P-wave velocity and unconfined compressive strength are the most effective variables and depth of coring and porosity are the least effective ones on the elastic modulus in the BPNN and RBFNN models, respectively. The main conclusion of this study is developing the new models with high accuracy and satisfied confidence in which all of the possible effective parameters were considered in estimation of the elastic modulus.
