Observed airborne 7Be activity concentration dependence on atmospheric variables is studied with the aim of characterising the behaviour of this radionuclide. Time series corresponding to 24 sites covering a significant range of longitudes and latitudes across Europe are modelled using the random forest algorithm. The Kolmogorov-Zurbenko filter is applied to each time series in order to separate the clean main signal from its noisy fraction. The clean fractions are then used as the target of the machine learning models, which are fed with surface and upper atmosphere variables from the ERA5 reanalysis and observed daily sunspot number time series. The validation of the models is addressed in terms of coefficient of determination ( \(R^2\) ) and Mean Absolute Error; a process in which most of the generated models appear to be representative enough to allow a significant feature importance analysis. The relevance of the features is divided into variables related to the 7Be cosmogenic generation mechanism, its atmospheric precipitation mechanism and the autocorrelation of the 7Be time series.