The recovery of nitrogen and phosphorus from simulated wastewater in the form of struvite was investigated through a Machine learning (ML)-based approach. The Extreme Gradient Boosting Algorithm (XCBoost) and Random Forest (RF) models were used for single-objective and multi-objective prediction of the recovery rates of N and P, respectively. The effects of seven process conditions on struvite crystallization were identifed. The results showed that XCBoost outperformed RF in both single-objective (R2=0.91-0.93) and multi-objective ( R2 =0.89) predictions. Furthemore, experimental validation was conducted with initial phosphorus concentrations of 10 mg/L and 1000 mg/L to determine the optimized process conditions for struvite recovery using the multi-objective model. The optimal conditions were found to be：N : P ratio of 1.2 : 1, Mg : P ratio of 1 : 1, pH of 9.5 , reaction time of 80 min, reaction temperature of 25℃ , and stirring rate of 240 r/min.