We apply a transformer-based model to classify the levels of Ozone in center of Madrid (Plaza España station). We also evaluate the performance in the same task of other machine learning algorithms such as Multi Layer Perceptron Netwroks, Long-Short Term Memory networks and Random Forest.

Data was extracted from public sources (http:https://www.aemet.es/ and https://datos.madrid.es/portal/site/egob). Data was previously preprocessed. It consisted firstly in the imputation of the previous value in missing values and in outlier values (3Q + 3IQR and 1Q - 3IQR). Then, we unified all data in an hourly format. The predictor variables used were:

-Hourly NO concentration in Plaza España Station.

-Hourly NO2 concentration in Plaza España Station.

-Hourly NOx concentration in Plaza España Station.

-Hourly CO concentration in Plaza España Station.

-Hourly SO2 concentration in Plaza España Station.

-Hourly O3 concentration in Plaza España Station.

-Daily rain in El Retiro Station (the closest meteorological station to Plaza España station).

-Daily maximum temperature in El Retiro Station.

-Daily minimum temperature in El Retiro Station.

-Daily average temperature in El Retiro Station.

-Daily maximum pressure El Retiro Station.

-Daily minimum pressure El Retiro Station.

-Weekday.

-Type of day (diary day, saturday, sunday, public holyday).

The target variable was:

-Hourly O3 concentration in Plaza España Station.