Over the past few years, the increase in the occurrence of floods has been primarily caused by climate change, and extreme meteorological conditions have significantly increased the vulnerability of the riverine bridges. During floods, the rapid flow of water may erode the soil (known as scouring, which is a primary contributor to bridge collapse) and reduce the lateral load-bearing capacity of the structure. In this regard, flood fragility curves have emerged as an effective tool to evaluate bridge vulnerability, often requiring a large amount of computational time and a complicated finite element (FE) model to adequately capture the predictor parameter space. Therefore, it is necessary to develop flexible parameterized demand models for riverine bridges conditioned on material parameters, loading parameters, and scouring parameters. This study develops demand metamodels for riverine bridges using machine learning algorithms. For this purpose, a detailed 3D finite element model of bridge piers with a pile foundation is developed, and nonlinear simulations are carried out considering uncertainty in material, flood loading, and scouring parameters. Various machine learning algorithms are employed to model the response of the bridge as a function of velocity and associated input variables. Goodness-of-fit measures are utilized to evaluate the predictive efficiency of the selected machine learning (ML) models. The results indicate that the predictive performance of the random forest model is considerably more accurate than the polynomial response surface model. Furthermore, the random forest approach yields the most precise predictions, exhibiting a high Adjusted R2 value with low symmetric mean absolute percentage error and root mean square error for bridge component responses.