BCC-based refractory complex concentrated alloys (RCCAs) are attracting attention as high-temperature materials because of their exceptional strength at high temperatures, but suffer from low tensile ductility. To search for alloys with improved ductility, it is necessary to investigate the properties of RCCA systems thoroughly, however, an experimental investigation of these vast constitutional spaces is impractical. This study employed data-driven approaches that combined first-principles calculations with machine learning. We first calculated the lattice parameters and elastic constants of 1693 ternary RCCAs, subsets of RCCAs alloys consisting of Ti, Zr, Hf, Nb, Mo, V, Ta, and W, using the exact muffin-tin orbitals method with coherent potential approximation (EMTO-CPA), and generated ductility-related parameters, including Pugh's ratio, Poisson's ratio, and Cauchy pressure. Machine learning models that could predict the three parameters were searched and trained using the generated data. Subsequently, an inverse design based on optimization algorithms was performed to identify optimal alloy systems with high Pugh's ratios. The ductility of the searched alloys was verified by calculating Pugh's ratio using EMTO-CPA, followed by thermodynamic calculations to investigate their structural stability.