Accurate modeling of a hysteretic behavior is essential for predicting various responses of structures subjected to random excitations. To this end, engineers have developed many approaches including mathematical modeling including the Bouc-Wen class models and data-driven approaches such as deep learning-based predictions. However, many existing models provide deterministic predictions, i.e., without quantifying model uncertainties. In engineering practice, such absence of uncertainty quantification may hamper engineer’s ability to make proper risk-informed decisions. This issue is especially critical in hysteresis prediction, where errors can accumulate as the analysis progresses over a time history. While there exist some probabilistic hysteresis models developed using Bayesian filters, the uncertainty of deep learning-based hysteresis models has not been investigated. The aim of this study is to develop a deep learning model capable of quantifying its model uncertainty, providing a measure of prediction credibility. In particular, deep learning models that are simply data-driven and those incorporating physics-based techniques are developed to compare their prediction uncertainties using a test dataset. This comparison of model uncertainties is expected to promote a proper model selection process minimizing the risk caused by model uncertainties. REFERENCES [1] Olivier, A., Shields, M. D., & Graham-Brady, L., Bayesian neural networks for uncertainty quantification in data-driven materials modeling, Computer Methods in Applied Mechanics and Engineering, 386. https://doi.org/10.1016/j.cma.2021.114079, 2021. [2] Strachan, R. W., & van Dijk, H. K., Bayesian Model Selection with an Uninformative Prior, Oxford Bulletin of Economics and Statistics, 65(SUPPL.), 863–876. https://doi.org/10.1046/j.0305-9049.2003.00095.x, 2003.