A comparative study of various hybrid neural networks and regression analysis to predict unconfined compressive strength of travertine

In this paper, the relationships between engineering properties of travertine rock samples including uniaxial compressive strength, density, Brazilian tensile strength and compressional and shear wave velocities were evaluated. The Bukan travertine mine located in Iran was considered as case study here. Various data analysis approaches including simple regression method, multiple regression method and artificial neural network (ANN) have been used for finding optimum estimation model for uniaxial compression strength of travertine rocks. Rock sample preparations difficulties and conducting expensive tests such as UCS motivated many researchers to study different regression methods to estimate UCS from other rock mechanic tests. In this paper, different statistical methods as well as some ANN optimization algorithms that were used by several researchers are compared to find the optimum solution for UCS estimation problem of travertine rock samples. These optimization tools comprising genetic algorithm, particle swarm optimization and imperialist competitive algorithm were applied to improve the efficiency of ANN analysis. Finally, after comparing all of the presented methods, the best results were obtained by ANN-PSO algorithm.