Forward osmosis (FO) systems have conventionally been investigated through analytical and physics-intensive modeling frameworks. In recent years, there has been a surging interest in understanding FO systems via ‘black box’ data-driven models based on machine learning (ML) principles. Typical ML studies on FO are biased toward computationally expensive models like artificial neural networks (ANNs), possibly due to their better prediction accuracies. In this study, the authors assess the potential of a less explored ML model, viz., Random Forest (RF), in predicting a crucial FO performance indicator, namely permeate flux. The authors then compare the flux predictions obtained from the RF model against a multiple linear regression (MLR) model and the traditional 1D solution–diffusion model incorporating the internal and external concentration polarization phenomena. The models are validated against experimental flux data available in the literature. It is shown that thoroughly optimized RF models may outperform facile implementations of expensive algorithms like the ANN. Our optimized RF model performs well with 92.2% prediction accuracy on unseen data, compared to the poorly performing MLR with 52.4%. However, both RF and MLR models lag in delivering prediction accuracies close to those obtainable from the 1D solution–diffusion model. It is concluded that while machine learning models yield decent predictions of FO performance, larger training datasets and well-optimized hyperparameters might be needed for them to outperform comprehensive physics-intensive models.