Alkali-activated concrete materials (AACMs) play an important role in decarbonising the construction industry. With the development of alkali-activated concrete from a versatile range of mineral precursors, a fundamental and holistic understanding of their durability performance is vital. Unlike blended Portland cement, the carbonation processes in AACMs differ significantly due to the absence of portlandite among the reaction products. Therefore, understanding, evaluating, and predicting the carbonation performance of these materials is essential. While laboratory testing for carbonation performance is accurate, it is also time intensive. In contrast, data-driven models provide an opportunity for quick diagnostic assessments. Previous studies have demonstrated the robustness of machine learning approaches in making sufficiently accurate predictions for mechanical properties. These modelling methods also allow for the assessment of current datasets, the identification of optimised mix design solutions for specific applications, and the exploration of complex correlations of key parameters that cannot be identified through traditional laboratory analysis. This study compared and evaluated the performance of different machine learning models for predicting the carbonation rate of alkali-activated concrete, using data from both laboratory tests and the literature. Primary factors controlling the chemical reactions between the AACMs binders, and the carbonation conditions were used as the input parameters. Predictive models using a range of machine learning algorithms, including Random Forest (RF), Gradient Boosting (GB), XG Boost (XGB), Artificial Neural Networks (ANNs), and Gaussian Processes (GP) were developed, and their prediction accuracy and robustness were assessed. The most representative model for alkali-activated concrete materials was identified and proposed. Among the evaluated models, GP demonstrated the highest prediction performance for carbonation depth.