Using recycled aggregates for substituting natural aggregates to prepare recycled aggregate concrete (RAC) has been regarded as an effective way to realize the sustainability of the construction industry. To promote RAC technologies in cold regions, it is necessary to ensure its frost durability. In this paper, three types of soft computing methods, including artificial neural network (ANN), Gaussian process regression (GPR), and multivariate adaptive regression spline (MARS), were used to model the frost durability of RAC based on the durability factor (DF) value. The results show that all the proposed models can predict DF values in good agreement with the experimental results. Compared with the GPR model and the MARS model, the ANN model showed the highest prediction accuracy. Based on the sensitivity analysis, the air-entraining type of RAC was the dominant variable that influenced the output for the experimental dataset and the three developed models. •Innovative prediction models for the frost durability of RAC were proposed.•A comparison between the performance of proposed models was studied.•The sensitivity analysis of inputs on the output for proposed models was performed.