Critical Air Quality Variables: A Data Mining Approach

Abstract

: Air quality monitoring is widely implemented worldwide; however, most existing approaches focus on measuring pollutant concentrations rather than identifying the minimal set of variables that most strongly influence air quality dynamics. This limitation increases system complexity and operational costs. Therefore, this study aims to identify the most influential environmental variables affecting air quality, represented by fine particulate matter (PM₂.₅ ) levels and classified into three states: Good, Moderate, and Poor. To do this, a database from the Science Data Bank was used, which includes: a) atmospheric pollutants (carbon monoxide (CO), sulfur dioxide (SO₂), nitric oxide (NO), nitrogen oxides (NOₓ) and tropospheric ozone (O₃)) and b) meteorological conditions (air temperature (TA), relative humidity (RH), atmospheric pressure (PA), dew point (DP), net radiation (NR), solar radiation (SR), direction (DIR) and average wind speed (VEL)). Using the J48 algorithm on the WEKA platform, complemented with a machine learning model developed in Python, a selection and prediction process of the dependent variable was carried out. The results show, with an accuracy of 94%, that the most influential variables in air quality are: a) pollutants: CO, Nox/TA, NOx, O₃ and b) meteorological: Air temperature (TA). These findings simplify environmental modeling and provide a replicable methodological basis for predictive air quality systems in different urban contexts.