Abstract: In order to improve the prediction accuracy of ρ(PM_2.5) and ρ(O₃) in Xi'an and better serve the prediction and warning work of Xi'an, based on the prediction results of the CAMx model, combined with the mesoscale WRF weather prediction data, ρ(PM_2.5) and ρ(O₃) observation data, this study optimized the simulation results of ρ(PM_2.5) and ρ(O₃) in Xi'an in 2019 based on multiple linear regression, ridge regression, lasso regression, decision tree, random forest and support vector machine model. The results showed that: (1) The CAMx model had bias in the prediction of pollutants, and the optimization model could obviously correct the systematic deviation of the CAMx model and improve the prediction accuracy. (2) The RMSE values of ρ(PM_2.5) and ρ(O₃) decreased from 174.00 and 37.11 μg/m³ to 34.36-39.37 and 24.77-28.82 μg/m³, respectively. The R values increased from 0.63 and 0.78 to 0.70-0.78 and 0.83-0.88, respectively. (3) Different models had different advantages in correcting the simulated values. The random forest model had a significant effect on ρ(PM_2.5) optimization, with an optimization improvement rate of 80%. The support vector machine model had the best effect on ρ(O₃) optimization, and the optimization improvement rate was 36%. The linear regression method had good optimization effect on ρ(O₃), but poor optimization effect on ρ(PM_2.5).The research results show that the machine learning algorithm has significantly optimized the CAMx simulation results, reflecting the research significance and feasibility of the machine learning algorithm to modify the results of the air quality numerical forecast model.