Abstract: This study evaluated the performance of Gradient Boosting Trees (XGB) and Deep Neural Networks (DNN) algorithms for air quality prediction in Jilin Province using meteorological forecasts and pollution observation data.The forecast evaluation results for 2023 demonstrated that the XGB model exhibited excellent trend-fitting capability in predicting pollutant concentrations.Its correlation coefficients for PM_2.5 and ozone concentration predictions generally exceeded 0.78 and 0.83,respectively,and the accuracy rate of AQI prediction ranges reached 40%-59%.It effectively captured high-concentration events but exhibited limitations in error control during stable pollution periods.While the DNN model captured the overall pollutant concentration trends,its overall performance was inferior,characterized by peak underestimation and unstable results.To address these limitations,the Optimized Combination Forecasting (OCF) algorithm was introduced to integrate the strengths of both models.The results indicated that OCF significantly enhanced prediction stability and accuracy,particularly during low-concentration pollution and stable periods.Compared to the standalone XGB model,OCF reduced the mean prediction bias for PM_2.5 and ozone by 1.4 μg/m³ and 7.1 μg/m³,respectively,through its ensemble strategy.In conclusion,while XGB delivered superior performance in most areas,while OCF demonstrated greater robustness,providing multi-dimensional technical support for refined quantitative air quality forecasting in Jilin Province.