Building performance simulation for energy retrofit of an existing building is a difficult task due to vast design options that lead to exhaustive iteration for vast design space. This study proposes an inverse approach using a machine learning model and Bayesian inference that automatically find feasible ranges of significant design parameters given a energy performance target. Design parameters can be chosen based on sensitivity ranking generated by the Sobol method, one of the global sensitivity analysis methods. The findings of this case study is illustrated in the following parallel coordinates chart. The proposed approach can be beneficially used for designers and engineers to make a time-efficient rational decision.