The durability evaluation of concrete materials based on the related experiments has a low economic efficiency ratio. The prediction accuracy of the conventional empirical formula for durability is restricted and the proportions of concrete mix cannot be calculated according to the performance. It is thus necessary to develop a novel and efficient material quality control and performance prediction tool. In this review, the process of building machine learning models was stated. The basic working flow and advantages of common algorithms, as well as the durability index prediction algorithms based on machine learning were summarized. Its application effects and development direction were discussed. This review can provide a basis for the in-depth development and application of machine learning technology in the field of concrete.