Alkali-activated lunar regolith is considered one of the most promising lunar regolith-based construction materials for large-scale lunar construction. This study presents a comprehensive framework integrating machine learning (ML) algorithms to investigate the influence of various features on the mechanical strength of alkali-activated lunar regolith simulant (AALRS), aiming to achieve the strength prediction and optimization design of AALRS. The properties of lunar regolith simulant, mixture proportions, preparation parameters, environmental conditions, and enhancement methods were employed as input features for ML modeling. The compressive and flexural strength predictive models were constructed using eight ML algorithms and evaluated through statistical indicators. Among the models, Extreme Gradient Boosting demonstrated the best performance, yielding an R² of 0.8684, a root mean square error of 6.2007 MPa, and a mean absolute error of 4.0874 MPa on the testing dataset. Using the best prediction models, four design strategies with three objectives, namely, compressive strength, flexural strength, and transport payload (TP), were optimized and evaluated using the non-dominated sorting genetic algorithm II and the technique for order preference by similarity to ideal solution methods. The proposed prediction and optimization framework for the mechanical performance of AALRS, which integrates extreme environmental effects and TP-driven design, provides a robust data-driven approach for the design, prediction, and optimization of AALRS, advancing the development of extraterrestrial construction materials and technologies.