As amply known, ozone concentration in the coastal area of study is well relevant in connection with “photochemical smog”, due to high levels of solar radiation and temperature values and possible photochemical oxidation of volatile organic compounds (VOCs) in the presence of nitrogen oxides (NOx). In this paper, a framework for predicting ozone concentration in urban area is presented, relying a LightGBM algorithm for gradient boosting on decision trees. The system represents a pragmatic and scientifically credible approach to data driven modelling applied to complex and uncertain situations. The study concerns the application of data analytic standard methodologies to air quality analysis, which includes the pre-treatment of data, the choice of a suitable configuration of the learning algorithm, the identification of the fitting parameters and error minimization. Training and verification data are significant statistical time-series over the past years validated from the air quality monitoring network in the urban area of Genoa (Italy).
Keywords: air quality, data driven model, machine learning, ozone, environmental quality.