Existing models for Radio Frequency (RF) and Free Space Optical (FSO) attenuations, such as the recommendations published by the International Telecommunication Union (ITU), require physical parameters along the communication channels. In practice, the weather parameters of the entire path are usually unavailable. This paper presents RF and FSO attenuation models built using machine learning algorithms and applied to empirical data. The empirical data consists of weather parameters collected at one end of the channel. Seven pairs of RF/FSO models are trained for specific weather conditions. The importance of each weather parameter is compared. RF attenuation is found to be sensitive to humidity, while FSO attenuation is closely related to scintillation. This paper shows how to obtain a pair of generic random forests that are applicable to seven specific weather conditions. The generic random forests predict the RF and FSO attenuations which have a joint distribution similar to the empirically observed distributions. They preserve the correlation between the RF and FSO attenuations as measured by the correlation coefficient and mutual information. When applied to empirical data, the generic random forests outperform the ITU models and models constructed by linear regression with interaction, both in terms of Root-Mean-Square-Error and R-squared.