Abstarct: Abstract Given the increasing use of concrete materials in the world and the daily performance of various tests on the types of concrete used in construction workshops, it is important to use networks capable of predicting the properties of concrete, more than that. Of the main advantages of this work is the time and money saved for engineers. Furthermore, by using prediction tools, it is possible to work out the optimum mixing plan and gain insight into the effects of various factors. Considering the variety of concrete types, it was decided to choose the latest concrete types, namely engineered cementitious composites, as the main subject of this research. ECC, which is another name for concrete that bends, is known to be able to strain due to the fibers present and the micromechanical design. Despite the fact that there are important factors that influence this ability, 13 have been used so far. Selected influencing factors such as the concrete mix design, e.g., water to cement ratio, weight percent of slag, fly ash, silica fume, sand, limestone powder, properties of fibers used in the cement composite, e.g., fiber content, fiber ratio, fiber tensile strength, fiber oil coating and other things include conditions and time of concrete storage, e.g., temperature, duration of curing in water and age of specimen. To advance the nose, a databaxse of laboratory data from the direct tensile test performed on the ECC specimen is needed in the first phase. By studying various articles, 469 groups of laboratory data were collected and classified, then were used for modeling. Soft computing models have also been used in the past as a tool for predicting the mechanical properties of concrete and have provided good results. In this study, the tensile behavior of ECC was predicted using the unique features of these types of algorithms. The best known of them are artificial neural network, neuro-fuzzy systems and the group method of data handling, which are also the models used here. Modeling is done in the three ways mentioned above, and then the best structure for each model is presented through the right and wrong method and validation factor checks. In the following, the final network is validated using a limited number of experimental data. As the results show, the error rate of the neural network is lower than that of the neural fuzzy method and is also mastered by the group method. In the last step, we can use the information from the artificial intelligence network to check the effects of the selected inputs on the target output. The results show that the temperature, the weight percentage, the fly ash and the sand have the greatest influence and the age of the sample has the least influence.