In probabilistic seismic hazard analysis (PSHA), the standard practice is to select a set of appropriate ground motion prediction equations (GMPEs) and assign weights on the logic tree, especially for regions where strong motion data are sparse and where no indigenous GMPE exists. Subjectively assigning weights to a set of models usually has the disadvantage of not obtaining mutually exclusive and collectively exhaustive models because of sparse or unavailable data. Therefore, the development of logic tree weightings in PSHA remains a major challenge. In this study, a distance metric measure for GMPEs is first analyzed to show how a set of GMPE’s prediction models can be partially reconciled by using high-dimensional information visualization techniques. Visualization of a large suite of GMPEs onto a 2-D graphical map provides a powerful theoretical framework that can guide the selection of a set of representative models. These models are considered mutually exclusive and collectively exhaustive, and have the ability to represent the center, body, and range of ground-motion distribution in a logic tree analysis. Second, determination a set of weights for PSHA are estimated based on the residuals, likelihood and EDR-index which are known as data – driven weight. The other weight type is non-data driven which is calculated based on the probability of a random model generated on a Sammon’s map. The methods presented here, that improve consistency in the weight assignment, can help to reduce overall epistemic uncertainties and offer a way of assigning weight on the logic tree. Finally, an example for assigning the weights on the logic tree for PHSA are discussed.