In recent decades, the World Health Organization and the scientific community have provided convincing evidence of the harmful effects on health due to exposure to atmospheric particles with a diameter of fewer than 10 microns or PM10. For this reason, this organization has alerted governments to measure, control, and evaluate air quality in real-time and in the medium term. For this reason, this work predicts PM10 particles using time series and a multilayer perceptron system to estimate PM10 concentrations over time. Four-time series models were analyzed, and four different multilayer perceptron architectures were used to determine the best prediction and know the pollutant's in short-term and medium-term future behavior. The results show that the neural system provides a better forecast in the short term with an error of 7.68% compared to the best time series method, the TIE, whose error was 27.20%. While in the medium term, the error was 31.10% for the TIE and 9.16% for the neural network.