Although in the last decade many indices and metrics have been developed to predict glare in the built environment, there is a lack of consensus-based glare metrics that could measure the level of visual discomfort indicated by people in their actual working environment. We present a new approach which integrates the use of Machine Learning algorithms for developing more accurate glare predictive model for open-plan offices. Field data were collected in four open-plan offices in Brisbane, Australia to obtain glare responses from 80 office workers and HDR images of their field of view using a calibrated smartphone [1]. We computed 24 glare metrics for each image and compared them with glare responses of the office workers. Using ROC analysis, we calculated the best thresholds for each glare metric based on the data collected which are only valid for an open-plan office building typology. We tested 825 Machine Learning models with various methods of training features to identify the best predicative ML model. We found that ML algorithms were better in finding the non-linear relationship between the input predictors (i.e. various multi-region luminance values, luminance, illuminance and glare indices) and the discomfort glare using the MRL method [2]. By comparing ML with ROC analysis, it was found that ML outperformed the conventional statistical methods with an overall accuracy of 83.8% (0.8 TPR, and 0.86 TNR) based on our dataset. Finally, we will provide the machine learning framework in the form of stand-alone software as shown in Figure1 which not only allows the users (architects, engineers, lighting designers) to use the trained model to predict glare, but also to train and develop new models using new HDR images [3].