Accurate glare prediction is a challenging task due to its subjective nature, as occupants' sensitivity to visual discomfort may differ under the same conditions depending on the individual. Although several glare metrics have been developed, limitations still exist. In this study, a new framework was proposed for developing new discomfort glare prediction models with higher accuracy using machine learning (ML) techniques. Machine learning algorithms act as classifier models which suit the stochastic nature of input data and non-linear problems. They were applied to explore their applicability in predicting glare in open-plan office spaces. Post occupancy evaluation (POE) data collected from 80 occupants in four open-plan offices in Brisbane, Australia were used to train ML models and test their prediction accuracy. Various glare indices and multi-region luminance values were extracted from high dynamic range images and used as input features for the training process. Results showed a high potential of using ML, as an overall accuracy of 83.8% was reached using the RUSBoost tree algorithm. The workflow was compiled into an easy-to-use tool called Open-Plan Glare Evaluator (OGE) for open-plan spaces with low illuminance conditions. The same workflow can be reapplied using more data collected from other locations to develop a global glare prediction model using machine learning.