Flotation procedure is a combination of many sub-processes which make its modeling quite complicated. Therefore, it is essential to use a method that can identify the most explanatory variables (feature selection) before modeling. Random forest (RF) with its associated variable importance measurements (VIMs) is an intelligent tool that has many advantages over other typical modeling methods This study investigated the effect of various flotation variables (particle characteristics: size (d1), and circularity (Cp) and hydrodynamicconditions: bubble Reynolds number (Reb), energy dissipation (ε), and bubble surface area flux (Sb)), on the flotation rate constant “k” and recovery “R” by VIM of RF, and predicted them based on the selected variables by RF models. VIMs indicated that the most effective variables for the k and R prediction were Sb-Reb-ε and d1-Cp-Sb, respectively. The predictive models yield satisfactory results for k and R with R2 = 0.96 and 0.97, respectively which demonstrate the robustness of RF as a prediction tool. These outputs verify that RF model can be used for feature selections and model developments within various complicated systems in mineral processing and separation techniques.