In order to effectively predict urban air quality, promote urban air pollution prevention and control, and make up for the deficiency of low accuracy and poor fault tolerance of traditional statistical models for urban air quality prediction under the background of big data era, a prediction model of urban air quality based on Stochastic Forest regression is proposed. Considering the pollutant concentration, meteorological parameters, time parameters and other factors, the optimal combination of parameters was adjusted by grid search method, and the urban air quality prediction model based on Stochastic Forest regression algorithm was established. Based on the index data of Chongqing from January 1, 2017 to July 31, 2020, the air quality in Chongqing is predicted and analyzed. The results show that the certainty coefficients of training set and test set are above 99%, and the mean square error and average absolute error under the model on the training set and test set are within the acceptable range, which proves that the model has the advantages of fast running speed, small prediction error, high prediction accuracy, and good learning ability and generalization ability.