This study proposes a methodology for online voltage stability monitoring using a feature subspace based ensemble approach. The overall idea is to use the input from varied feature selectors for the ensemble and aggregate their outputs. This approach is superior to conventional feature selection methods because it can handle stability issues that are usually poor in existing feature selection methods and improve performance. The selected features are used as an input to three different regression algorithms to enable online voltage stability monitoring. A Bayesian optimization technique is used to tune machine learning (ML) models’ hyper-parameters and determine the optimal number of features. The proposed approach is evaluated in experiments using simulated data from the Nordic test system. The simulation results have shown that the proposed method efficiently predicts the status of dynamic voltage stability in the test system.