Monitoring air quality is a topic of current interest, since poor quality has a negative impact on health. Air quality is affected by different pollutants, such as particulate matter and gases, produced by the growing industrial development. As a preventive measure, Mexico established different standards in order to control airborne pollution. In this paper, we propose a methodology based upon a recurrent long-term/short-term memory network for the prediction of exceedances of PM10 (particles of less or equal diameter than 10 micrometers) with time intervals of 72, 48 and 24 hours in advance. Obtaining a satisfactory percentage of prediction as a whole a minimum variability in repetitive experimental runs.