Free Space Optical Communication (FSO) technologies allow for high transmission rates without RF interference. However, FSO systems are vulnerable to atmospheric turbulence and other environmental factors that degrade the signal quality. Channel estimation is an essential part of optimizing FSO communication. However, traditional methods need to be more accurate and inflexible in terms of channel estimation. This study explores FSO channel estimation with Machine learning (ML). ML is capable of learning complex patterns and adapting to dynamic environment conditions. The goal of this study is to develop ML-based models to predict FSO channel characteristics. The approach includes collecting large-scale channel measurement data in actual FSO communication environments, and training and validation of ML models for accurate, dynamic channel estimation. The results of this study demonstrate that ML-based channel estimation offers substantial improvements in adaptability, as well as accuracy, compared to traditional methods. This can transform FSO communication systems by increasing their robustness and reliability in various environmental conditions.