Air quality assessment requires concentration measurement of various polluting gases, typically requiring multiple expensive sensors, each dedicated to a specific pollutant. Therefore, there is a need for cost-effective sensors capable of detecting one or more pollutants, with less reliability, such as camera-based sensors, but enabling denser sampling. In this paper, we investigate how the extinction coefficients estimated from an infrared camera may be useful for predicting ground-level ozone concentration. In addition to these coefficients, we show how weather and pollution measures collected from stations near the studied area are useful, with different machine learning methods, to better predict ozone concentration. The performance of the models is validated through a comprehensive evaluation using MAE, RMSE and R-Squared metrics. Parameter selection methods are also used to study the impact of different meteorological parameters and other pollutant concentrations on the prediction of ozone concentration.