Federated learning has seen a rise in interest given that privacy concerns of contemporary data can hinder data pooling. To this end, several machine learning algorithms including random forests have already seen use in some kind of federated learning setting. A recently developed variation of the random forest algorithm, called the evolutionary random subspace forest (ERSF), can be seen as a viable candidate in this field due to its highly parallelizable and iterative nature. This work assessed the potential of the ERSF to be extended to a federated learning algorithm using a simulated dataset representing a regression problem. The generated dataset was split and distributed amongst different nodes. A central node aggregated the locally trained ERSFs from each node into a global forest which was sent back out to the nodes for further local training. With a performance comparable to that of an ERSF trained on pooled data, the federated learning algorithm was able to outperform ERSF's which were trained fully separate on each node. Comparison with the forests trained fully separate also showed the current algorithm to be less prone to overfitting. This reaffirms the advantages of the federated learning paradigm and establishes a promising framework for future research.