With the expansion of urban enviroments, new challenges are raised for the excavation and stabilization of deep excavations. One of the factors in the safe and economic design of retaining structures, is the phenomenon of soil arching. Considering soil arching phenomenon, it is possible to reduce the construction costs of retaining structures. At the first step to evaluate effective parameters in soil arching of retaining structures that are combined of steel piles of type (2PIE300) and steel anchores, PLAXIS 3D TUNNEL software is used to simulate 3D numerical modeling. For better comparison of results, the external loading conditions and technical specifications of structural elements are assumed to be identical. In order to determine effective parameter ranges of the fine (CL-ML) to coarse soil (SC-SM), soil characteristics of stations A2 to L2, Line 2 of Mashhad subway, the model with hardening soil behavior (HS) has been used. Modeling started with 2 meters horizontal ,vertical distance and increased to 4 meters. Then , the relationship between safety factor and arching was discussed. In order to provide a prediction model; internal friction angle of soil, soil density, distance between anchors and elastic modulus were chosen based on previous studies. Other parameter values except the elastic modulus ??were used in the analysis of multiple regression and modeling with artificial neural network. The results showed that the best regression model can be derived when correlation between measured and predicted values ??is equal to 94%. Also correlation coefficient obtained from the neural network that is equal to 99%, showed that the neural network is a powerful tool in understanding the relationships between variables and predicting the safety factor.