Free space optical communications (FSO) offers many and varied advantages, including high data transfer speed, ease of installation and deployment compared to fiber-optic communications, in addition to not requiring a license. However, this technology suffers from some obstacles, which are represented by weather conditions that obstruct the laser during the transmission process. In addition, in recent years, artificial intelligence technologies have been introduced into free-space communications systems. This research presents work in two stages. The first is to implement a simulation of the FSO system using Optisystem program with two types of modulation, Non Return to Zero (NRZ) and Return to Zero (RZ), in different weather conditions (Clear – Rain – Fog). In the second stage, machine learning algorithms have been used to predict the best type of modulation depending on the weather condition. The results showed that the Random Forest (RF) algorithm gave better results in prediction accuracy compared to the Support Vector Regression (SVR) and K-Nearest Neighbors (KNN) algorithms.