Models for forecasting the concentration of atmospheric pollutants have become a fundamental tool in air quality management. In this study, the precision in predicting hourly surface concentrations of tropospheric ozone (O3) and carbon monoxide (CO) was evaluated between a deterministic model (WRF-Chem) and two statistical models (Artificial Neural Networks (ANN) and support vector regression (SVR)), using the city of Manizales, Colombia as a case study. The results obtained show that the statistical models notably improve the performance of the predictions made by the deterministic model at the evaluated point. For O3, the ANN model improved the correlation values ​​(R: 0.89 vs 0.70) compared to WRF-Chem, just as it will lose the reduction in the deviation of the predictions (MB: -0.16 vs 3.10, RMSE: 2.33 vs 5.95). Likewise, the SVR model improved the different performance statistics for CO compared to the deterministic WRF-Chem model (R: 0.68 vs 0.47, MB: -0.01 vs -0.23, RMSE: 0.25 vs 0.42). The best performance offered by the statistical models is explained because they were developed from information obtained locally, allowing the characteristics of the study area to be adequately captured. This study demonstrates the applicability of statistical models as an economic and easy-to-implement tool in cities of emerging countries, being useful for forecasting and filling in data.