Fluid-Structure Interaction (FSI) simulations present significant computational challenges for Uncertainty Quantification (UQ) due to expensive coupled multiphysics solvers and time-dependent responses. When a single FSI simulation requires 1-2 hours, the traditional Monte Carlo approach requires thousands of evaluations, making UQ computationally prohibitive. This work presents a comparative study of established surrogate modeling strategies, namely Polynomial Chaos Expansion (PCE), Kriging (Gaussian Process Regression) and hybrid PCE-Kriging. By integrating OpenIFEM, a high-fidelity FSI solver, with Dakota UQ toolkit we create a flexible framework to test which method best handles the complexities of transient FSI problems by evaluating surrogate performance across multiple benchmarking problems. Our analysis reveals strongly problem-dependent performance characteristics : pure PCE achieves exceptional accuracy for problems with naturally polynomial physics such as beam deflection under distributed loads. Kriging more effectively captures local nonlinearities with continuous vortex shedding around cylinders. For problems combining smooth global trends with localized nonlinear effects, as in FSI cases, the hybrid PCE-Kriging approach offers the best balance. Initial findings show that Latin Hypercube Sampling with 60-70 training samples enables 16-20 times computational speedup over Monte Carlo while maintaining engineering-grade accuracy for global sensitivity analysis. This comparative study provides practical guidance for surrogate method selection to quantify the uncertainty of expensive, time-dependent FSI problems.