Recent advances in high-entropy alloys (HEA) broaden their application from structural materials with excellent combination of strength and ductility to nano-sized catalysts. In order to achieve the optimum performance in each application, it is important to fabricate designated phases in the alloy of interest. Combinations of 4 to 6 elements having high-melting temperature can form refractory HEAs, where BCC structure is favored due to its superior property in elevated temperature. In this work, a data-driven approach is proposed for the phase prediction in refractory HEAs based on the collected experimental data from the literature, CALPHAD computations, and data analyses. More than 300 sets of data on the different alloy compositions, melting and thermo-mechanical conditions, mechanical properties, and constituent phases confirmed by X-ray diffraction were collected from the literature. Thermodynamic properties such as liquidus and solidus temperatures, and enthalpy of mixing were added to extend relevant features. The constituent phases were categorized in 4 classes; (i) single BCC (ii) phase-separated BCCs (iii) BCC, FCC and/or HCP (iv) mixture of BCC and intermetallic. When the feature importance in the machine learning models were explained by Sharpley values, it was turned out that the most important feature on the phase prediction model is the alloy composition.