A short-term deep learning model for urban pollution forecasting with incomplete data

Abstract

A deep neural network model for the short term prediction of Ozone, 10 micrometers particulate matter and 2.5 micrometers particulate matter concentrations in a major northwestern metropolitan area of Mexico is developed. In order to formulate such a model, the data avail- able from the local air quality automatic network monitoring system are used for training, validation and testing purposes. Such time series data are incomplete and a procedure of missing data imputation is carried out. The model predicts with high accuracy the concentration of the target pollutants and the training procedure, performance metrics and tools used are discussed in this work. Such a model can be used for the implementation and evaluation of public politics for improving population health, and reducing the potential negative impacts of harmful pollutants by issuing early warnings on dangerous pollution levels.