The technology of recycled aggregate concrete （RAC） is an effective way to realize the resource utilization of construction waste. By collecting the data on the chloride ion corrosion test and carbonation test of RAC， variables related to materials and test environments were set as input parameters. The electric flux and carbonation depth were used to quantify the RAC’s resistance to chloride ion corrosion and carbonation， respectively. The machine learning methods were used to construct the prediction models of durability for RAC. On this basis， taking strength， durability， and cost as the optimization goals， the optimal design method of the RAC mixture was proposed by combining the NSGA-Ⅱ algorithm and the technique of order preference similarity to the ideal solution. Results show that the gradient boosting tree model can predict the RAC’s resistance to chloride ion corrosion better than other models， and the Gaussian process regression model has the best performance in predicting the RAC’s resistance to carbonation. A low-cost RAC mixture that meets the requirements of durability and mechanical properties was obtained with the proposed mixture optimization design method， which can be used to guide the construction mix design.