Free space optical (FSO) communication systems provide high-bandwidth data transmission and resistance to electromagnetic interference, making them a compelling choice for existing wireless optical networks. FSO System performance is influenced by environmental factors, channel length, weather conditions, and modulation techniques. This work examines the application of a random forest classifier for the implementation of an adaptive modulation system to enhance performance under fluctuating channel circumstances. Bit error rate (BER) and received signal power for different link distances and weather conditions for phase shift keying (PSK) modulation schemes were observed from the simulation. The random forest classifier was trained on this data to predict the appropriate modulation strategy for a specific channel state. The findings indicated that the random forest-based adaptive modulation system significantly reduced BER and enhanced signal integrity under various weather conditions and link distances. Analysis of feature importance indicated that link distance and weather conditions were the predominant factors in ascertaining the appropriate modulation strategy. This study highlights the capability of machine learning methods to improve the resilience and adaptation of FSO communication systems in variable situations.